Method for identifying whether an obstacle object is present in a detection area

Abstract

The invention relates to a method for operating an ultrasonic sensor, to a sensor thereof, to a data transmission to a computer system controlling the ultrasonic sensor and to a decompression method in the computer system. The compression method comprises the following steps: detecting an ultrasonic reception signal of an ultrasonic transducer; providing a signal-free ultrasonic echo signal model (610); at least once, optionally repeatedly, performing the following steps: subtracting the reconstructed ultrasonic echo signal model (610) from the ultrasonic reception signal (1) and forming a residual signal (660), performing a method for identifying signal objects in the residual signal (660); supplementing the ultrasonic echo signal model (610) with parameterized signal waveforms of the identified signal objects (600 to 605); and ending the repetition of the steps if the value of the residual signal is below a value of a pre-defined threshold signal. At least a part of the identified signal objects is then transmitted to the computer system in the form of a signature for the identified signal objects and at least a part of the identified signal objects is used there. Preferably, the signal objects are reconstructed in the computer system into a reconstructed ultrasonic echo signal model.

Description

[0001] This patent application claims priority to German patent application 10 2019 106 204.7, dated March 12, 2019, the contents of which are incorporated herein by reference and are the subject of this patent application. Technical Field

[0002] The present invention relates to a method for identifying obstacle objects based on signals from multiple sensors and predicting changes in the position of known obstacle objects, as well as for compressing and decompressing sensor signals used for the above purposes, particularly for vehicles, for detecting the environment of the vehicle, as provided, for example, in the field of vehicle parking assistance.

[0003] This invention particularly relates to a method and a corresponding apparatus for transmitting data from an ultrasonic sensor system in a vehicle, used for example parking assistance, to a computer system (hereinafter also referred to as a controller). The basic idea is to identify potentially important structures in the measurement signal and compress the measurement signal by transmitting only these identified, potentially important structures, rather than the measurement signal itself. The actual object, such as an obstacle in the parking process, is then identified after the measurement signal is reconstructed in the computer system, which typically aggregates multiple such decompressed measurement signals from multiple ultrasonic sensor systems. Background Technology

[0004] Ultrasonic sensor systems for parking assistance are becoming increasingly popular in vehicles. In this context, there is a growing demand for transmitting increasing amounts of the actual measurement signals from the ultrasonic sensors to a central computer system. This central computer system processes this data, along with data from other ultrasonic sensor systems and other types of sensor systems, such as radar systems, through sensor fusion to create what is known as the surrounding environment or an environmental map. Therefore, it is desirable to perform object recognition not within the ultrasonic sensor system itself, but first within the computer system through sensor fusion, to avoid data loss and thus reduce erroneous information, thereby lowering the probability of incorrect decisions and, consequently, accidents. However, the bandwidth of the available sensor data buses is limited. Replacing these sensor data buses should be avoided, as they have already been field-proven. Therefore, it is desirable not to increase the amount of data to be transmitted. In short: the information content of the data and its importance for subsequent obstacle recognition (object recognition) in the computer system must be enhanced, ideally without significantly increasing the data rate, and even better, without increasing the data rate at all. On the contrary: preferably, the data rate requirement should even be reduced to allow for the data rate capacity used to transmit status data and self-test information from the ultrasonic sensor system to the computer system, which is mandatory within the framework of functional safety (FuSi). This invention addresses this problem.

[0005] Various methods for processing ultrasonic sensor signals are known in the prior art.

[0006] A method for evaluating echo signals used in vehicle environmental detection is known, for example, from WO-A-2012 / 016834. This document proposes: emitting a measurement signal with a pre-defined encoding and format, and searching for and identifying portions of the measurement signal in the received signal by means of correlation with the measurement signal. Then, a threshold is used to evaluate the level of this correlation, rather than evaluating the level of the echo signal envelope.

[0007] As is known from DE-A-4 433 957: In order to perform obstacle identification, ultrasonic pulses are periodically emitted and the location of the obstacle is inferred based on the transit time, wherein, in this evaluation case, the echoes are made to remain correlated over multiple measurement cycles, while the uncorrelated echoes are suppressed.

[0008] A method for decoding a received signal received from an ultrasonic sensor in a motor vehicle is disclosed in DE-A-10 2012 015 967, wherein the transmitted signal of the ultrasonic sensor is transmitted in an encoded manner and the received signal is correlated with a reference signal for decoding, wherein the frequency shift of the received signal relative to the transmitted signal is determined before correlating the received signal with the reference signal, and the received signal is correlated with the transmitted signal, which serves as the reference signal, having been shifted with respect to the frequency of the received signal by the determined frequency shift, wherein in order to determine the frequency shift of the received signal, a Fourier transform is performed on the received signal and the frequency shift is determined based on the result of the Fourier transform.

[0009] A method for detecting the surrounding environment of a vehicle by means of ultrasound is known from DE-A-10 2011 085 286, wherein ultrasonic pulses are emitted and ultrasonic echoes reflected at the object are detected. The detection area is divided into at least two distance regions, wherein ultrasonic pulses for detection in the respective distance regions are emitted independently of each other and encoded with different frequencies.

[0010] A device and method for an environmental sensing apparatus using a signal converter and an evaluation unit are disclosed in WO-A-2014 / 108300, wherein a signal received from the environment having a first pulse response length at a first time point during a measurement period and a longer second pulse response length at a later second time point within the same measurement period is filtered according to the transit time.

[0011] A method for providing information, depending on obstacle objects detected in the surrounding environment of a motor vehicle, is disclosed in DE-A-10 2015 104 934. In this disclosed method, a sensor device of the motor vehicle is used to detect the surrounding environment and the information is provided at the vehicle's communication interface. Here, raw sensor data, as information about the free space identified between the sensor device and the detected obstacle objects in the surrounding environment, is stored in a control unit on the sensor device side. According to this disclosed method, this raw sensor data is provided at a communication interface connected to the control unit on the sensor device side for transmission to a processing device for constructing a map of the surrounding environment and for further processing by the processing device for constructing the map of the surrounding environment.

[0012] The technical teachings of the aforementioned patents are guided by the idea that ultrasonic sensors are used to identify objects in the vehicle's environment, and that object data is only transmitted after the object has been identified. However, the synergistic effect of using multiple ultrasonic sensors disappears in this case.

[0013] A method for detecting the environment of a vehicle using multiple sensors is disclosed in DE-A-10 2010 041 424, wherein at least one sensor detects at least one echo information about the environment during at least one echo cycle and uses an algorithm to compress the at least one echo information, and wherein the at least one compressed echo information is transmitted to at least one processing unit. While this known method also focuses on the compressed transmission of data of identified objects, it has been recognized that it is meaningful to compressively transmit data (echo information) extracted from the received echo signal to the controller without specifying a compression method.

[0014] A method for sensor connection is disclosed in DE-A-10 2013 226 373. A method for operating a sensor system, particularly an ultrasonic sensor system, having an ultrasonic sensor and a controller is disclosed in DE-A-10 2013 226376. The ultrasonic sensor transmits data to the controller in a current-modulated manner. The controller transmits data to at least one ultrasonic sensor in a voltage-modulated manner.

[0015] DE-A-10 024 959 discloses a device for one-way or two-way data exchange between a measuring unit for measuring a physical quantity and an adjustment / evaluation unit, the adjustment / evaluation unit determining the physical quantity with a predetermined accuracy based on measurement data provided by the measuring unit. The measuring unit and the adjustment / evaluation unit are connected via a data line. Data transmission between the measuring unit and the remote adjustment / evaluation unit should be optimized, and for this purpose, at least one compression unit is provided, which is assigned to the measuring unit and compresses the measurement data provided by the measuring unit within a predetermined period, such that, on the one hand, the information content of the measurement data is transmitted to the adjustment / evaluation unit via the data line in a manner that fully or minimally determines the physical quantity with a predetermined accuracy, while on the other hand, the amount of data transmitted is minimized. DE-A-102013 015 402 discloses a method for a sensing device for operating a motor vehicle, wherein a transmitted signal is sent to the surrounding environment area of ​​the motor vehicle by means of a sensor of the sensing device, particularly an ultrasonic sensor, and the signal component of the transmitted signal 11 reflected at a target object in the surrounding environment area 7 is received as an echo raw signal 12. Data is transmitted between the sensor and controller of this sensing device via a data bus. The sensing device is used to determine measurement parameters about the target object. The raw echo signal is converted into a digital echo signal using a sensor converter, and this digital echo signal is transmitted from the sensor to the controller via the data bus. The controller determines the measurement parameters by correlating the digital echo signal with a reference signal corresponding to the transmitted signal. However, data decompression within the controller is not provided in this configuration.

[0016] A system for sensor fusion is known from US-A-2006 / 0250297.

[0017] A method for efficiently transmitting data from an ultrasonic system to a data processing device via a vehicle bus is disclosed in WO-A-2018 / 210966. However, this disclosed method does not involve the implementation of object recognition.

[0018] Methods for transmitting data from an ultrasonic system to a data processing device via a vehicle bus are known in DE-B-10 2018 106 244 and DE-A-10 2019 106 190 (published after the priority date of this PCT application), but do not describe how obstacle object recognition is performed. Summary of the Invention

[0019] This invention relates particularly to various methods for compressing sensor signals, especially those of an ultrasonic measurement system for a vehicle, then transmitting these sensor signals from the sensors to a data processing unit, and for decompressing the transmitted compressed sensor signals in the data processing unit. It should be emphasized that compression is performed based on characteristic signal objects or waveform features of the sensor signals, without identifying, or being unable to identify, what type of obstacle object is located in the vehicle's environment based on these characteristic signal objects or waveform features. More precisely, according to another inventive proposal, which will also be described later, physical obstacle objects are identified within the data processing unit itself, i.e., based on the decompressed sensor signals from multiple sensors.

[0020] Another key aspect of this invention is predicting changes in identified obstacle objects over a predicted time period. This is based on the fact that in current methods of detecting the vehicle's surroundings using ultrasonic signals, there is a dead time during which measurements cannot be performed; instead, ultrasonic signals are emitted, which is applied to the ultrasonic transducer, receiving the echo signal after the attenuation phase. These dead times can be particularly long when the vehicle's surroundings to be inspected also have a long-range area, as encoded ultrasonic signals must be used, making the processing of these signals even more time-sensitive.

[0021] Therefore, the objective of this invention is to provide a solution that addresses the aforementioned objective of reducing the bus bandwidth required for transmitting measurement data from an ultrasonic sensor system to a computer system, and has other advantages in meeting the aforementioned other objective requirements.

[0022] According to the present invention, this task is solved by the method described herein.

[0023] To address this task, according to a first variant of the invention, a method for transmitting sensor signals from a transmitter to a data processing unit is proposed, specifically for applications in vehicles and particularly for transmitting ultrasonic sensor signals from an ultrasonic sensor to a data processing unit, wherein in this method,

[0024] - Provide sensor signals from sensors S1, S2, S3, ..., Sn.

[0025] - The sensor signal data describing the sensor signal is transmitted wirelessly or via wired connection from sensors S1, S2, S3, ..., Sn in a compressed manner to the data processing unit ECU.

[0026] - In order to compress the sensor signal data describing the sensor signal

[0027] - By using feature extraction (FE) to extract signal waveform characteristics from the sensor signal,

[0028] - The total feature vector signal F1 is formed from the extracted signal waveform characteristics.

[0029] - Assign signal objects to the features of the total eigenvector signal F1 in the following way:

[0030] a) Using the artificial neural signal object recognition network NN0, signal objects are identified for each feature.

[0031] b) For each identified signal object, a reconstructed single feature vector signal R1, R2, Rn-1, Rn is generated using another artificial neural single reconstruction network among multiple artificial neural single reconstruction networks NN1, NN2, ..., NNn.

[0032] c) Subtract the sum of all reconstructed individual eigenvector signals R1, R2, Rn-1, Rn from the total eigenvector signal F1 to form the eigenvector residual signal F2.

[0033] d) As long as the residual signal F2 of the eigenvector is less than a pre-given threshold signal, the assignment of signal objects to the features of the total eigenvector signal F1 is terminated.

[0034] e) Otherwise, in order to identify other potential signal objects, the residual feature vector signal F2 is fed into the neural signal object recognition network NN0.

[0035] f) For each of the other identified potential signal objects, generate reconstructed single feature vector signals R1, R2, Rn-1, Rn using one of the neural single reconstruction networks NN1, NN2, ..., NNn.

[0036] g) Subtract the sum of all the individual eigenvector signals R1, R2, Rn-1, Rn reconstructed in step f) from the total eigenvector signal F1 to form the updated eigenvector residual signal F2.

[0037] h) Repeat steps e) to g) until the residual signal F2 of the corresponding updated feature vector is less than a pre-given threshold signal, and

[0038] i) End the assignment of signal objects to the features of the total feature vector signal F1.

[0039] - For each identified signal object, generate signal object data representing that signal object, and

[0040] - Transmit these signal object data to the data processing unit ECU.

[0041] In this variant of the invention, sensor signal data is compressed by feature extraction from the sensor signal. Based on the signal waveform characteristics, a total feature vector signal (Feature-Vektor-Signal) is formed using this feature extraction. That is, the features of this vector signal correspond to the signal objects of the sensor signal. Here, an artificial neural total compression network is used to identify the signal objects for each feature of the vector signal. This neural total compression network is correspondingly trained and generates one or more signal objects. These signal objects are advantageously stored in a single memory and reconstructed into single feature vector signals using a set of artificial neural single reconstruction networks. Here, each of the aforementioned memory is allocated a neural single reconstruction network. Then, the sum of all reconstructed single feature vector signals is subtracted from the total feature vector signal, such that a residual signal of the total feature vector signal is retained. If the residual feature vector signal is less than a predetermined threshold signal, the process of assigning signal objects to features of the total feature vector signal ends. Otherwise, the above process continues iteratively.

[0042] Next, at the end of the above process, signal object data is generated, representing the identified signal objects. In terms of data volume, this representation is much smaller than when it is desired to transmit sensor signals, for example, in the form of sampled value signals. The sensor signal data, which is compressed in this respect, is then transmitted to the data processing unit via wired or wireless means.

[0043] The above process of identifying signal objects can be implemented by spatial or temporal multiplexing based on the characteristics of the total feature vector signal according to steps a) to i).

[0044] To determine the presence of certain signal waveform characteristics, the sensor signal is examined; specifically, in the case of an ultrasonic receiver or transducer, the electrical signal formed by converting the received acoustic ultrasonic signal is examined. Signal waveform characteristics include, for example, a signal level increasing from below a predetermined threshold to above that threshold (or vice versa), a local or absolute maximum or minimum value. Multiple such temporally consecutive signal waveform characteristics constitute signal objects. Signal waveform characteristics are typically stored as feature vectors, parameter values ​​generated through sampling, obtained through feature extraction. Now, based on the sequence of feature vectors that, in turn, form the feature vector signal, different sets of successive signal waveform characteristics can be identified as signal objects, where multiple such signal objects are temporally consecutive. In the case of an ultrasonic measurement system, these signal objects correspond to different real-world obstacle objects at different distances from the ultrasonic measurement system.

[0045] In the case of compressing sensor signal data (sampled values ​​of the sensor signal) that describes the sensor signal, according to the first variant, each successive signal object is gradually identified by a feature vector signal that includes all temporally successive signal objects, hereinafter referred to as the total feature vector signal. For each identified signal object, a single feature vector signal is reconstructed, representing that identified signal object. Now, for each identified signal object, this single feature vector signal is subtracted from the total feature vector signal, such that after a finite number of iterations, all signal objects in the total feature vector signal are identified, and a feature vector residual signal below a predetermined threshold is obtained.

[0046] According to a variation of the method of the present invention, the above process is performed iteratively, in that signal objects are identified separately from the total feature vector signal by means of a neural signal object recognition network. This signal object is typically the one that dominates in terms of corresponding information in the total feature vector signal, and is stored in a buffer. After inverse transformation into a single feature vector signal (either through a single neural reconstruction network allocated to this buffer or through a single neural reconstruction network common to all buffers), this single feature vector signal is then subtracted from the total feature vector signal, and the resulting feature vector residue is fed to the neural signal object recognition network, which then identifies the next signal object, which in turn also involves the signal object whose information is dominated by the corresponding current feature vector residue. This next signal object is then stored in another buffer. After inverse transformation into a single feature vector signal by a single neural reconstruction network allocated to this buffer, a single feature vector signal is formed, and this single feature vector signal is now subtracted from the feature vector residue. In this manner and method, all signal objects are gradually identified.

[0047] In this regard, the present invention proposes a method for transmitting sensor signals from a transmitter to a data processing unit in a measurement system, particularly for distance measurement and especially for applications in vehicles, and particularly for transmitting ultrasonic sensor signals from an ultrasonic sensor to the data processing unit of an ultrasonic measurement system, wherein in this method...

[0048] - Provides sensor signals from sensors S1, S2, S3, ..., Sm.

[0049] - The sensor signal data describing the sensor signal is transmitted wirelessly or via wired connection from sensors S1, S2, S3, ..., Sm in a compressed manner to the data processing unit ECU.

[0050] - In order to compress the sensor signal data describing the sensor signal,

[0051] - By using feature extraction (FE) to extract signal waveform characteristics from the sensor signal,

[0052] - The total feature vector signal F1 is formed from the extracted signal waveform characteristics.

[0053] - Assign signal objects to the features of the total eigenvector signal F1 in the following way:

[0054] a) Using an artificial neural signal object recognition network NN0, signal objects are identified based on the features of the total feature vector signal F1.

[0055] b) Store the data representing the identified signal objects in buffer memories IM1, IM2, ..., IMn.

[0056] c) For the identified signal objects, reconstructed single feature vector signals R1, R2, Rn-1, Rn are generated using a common artificial neural single reconstruction network NN1, NN2, ..., NNn, or using a corresponding artificial neural single reconstruction network among multiple artificial neural single reconstruction networks NN1, NN2, ..., NNn.

[0057] d) Subtract the reconstructed individual eigenvector signals R1, R2, Rn-1, Rn from the total eigenvector signal F1 to form the eigenvector residual signal F2.

[0058] e) As long as the residual signal F2 of the eigenvector is less than a pre-given threshold signal, the assignment of signal objects to the features of the total eigenvector signal F1 is terminated.

[0059] f) Otherwise, in order to identify other potential signal objects, the residual feature vector signal F2 is fed into the neural signal object recognition network NN0.

[0060] g) Repeat steps b) to f) until the residual signal F2 of the corresponding updated feature vector is less than a pre-given threshold signal, and

[0061] h) End the assignment of the signal object to the features of the total eigenvector signal F1, and

[0062] - For each identified signal object, generate signal object data representing that signal object, and

[0063] - Transmit these signal object data to the data processing unit ECU.

[0064] In another embodiment of the invention, the invention relates to a method for transmitting sensor signals from a transmitter to a data processing unit in a measurement system, particularly for distance measurement and especially for applications in vehicles, and particularly for transmitting ultrasonic sensor signals from an ultrasonic sensor to a data processing unit of an ultrasonic measurement system, wherein in this method...

[0065] - Provides sensor signals from sensors S1, S2, S3, ..., Sm.

[0066] - The sensor signal data describing the sensor signal is transmitted wirelessly or via wired connection from sensors S1, S2, S3, ..., Sm in a compressed manner to the data processing unit ECU.

[0067] - In order to compress the sensor signal data describing the sensor signal,

[0068] - By using feature extraction (FE) to extract signal waveform characteristics from the sensor signal,

[0069] - The total feature vector signal F0 is formed based on the extracted signal waveform characteristics.

[0070] - Assign signal objects to the features of the total eigenvector signal F0 in the following way:

[0071] a) Using an artificial neural signal object recognition network NN0, signal objects are identified based on the features of the total feature vector signal F0.

[0072] b) Store the data representing the identified signal objects in buffer memories IM1, IM2, ..., IMn.

[0073] c) For the identified signal objects, reconstructed single feature vector signals R1, R2, Rn-1, Rn are generated using a common artificial neural single reconstruction network NN1, NN2, ..., NNn, or using a corresponding artificial neural single reconstruction network among multiple artificial neural single reconstruction networks NN1, NN2, ..., NNn.

[0074] d) The reconstructed individual feature vector signals R1, R2, Rn-1, Rn are inversely transformed into signal objects, more precisely, in particular, the inverse transform IFE of the feature extraction FE of the sensor signals is used to inversely transform the reconstructed individual feature vector signals into signal objects.

[0075] e) Subtract the inversely transformed signal object from the sensor signal to form the residual sensor signal.

[0076] f) As long as the residual sensor signal is less than a pre-defined threshold signal, the assignment of signal objects to the features of the total feature vector signal F0 is terminated.

[0077] g) Otherwise, feature extraction (FE) is used to extract signal waveform characteristics from the residual sensor signal, and a total feature vector residual signal F0 is formed based on these signal waveform characteristics.

[0078] h) Based on the corresponding current total feature vector residual signal F0, repeat steps a) to g) until the residual sensor signal is less than a pre-given threshold signal, and

[0079] i) End the assignment of signal objects to the features of the total eigenvector signal F0, and

[0080] - For each identified signal object, generate signal object data representing that signal object, and

[0081] - Transmit these signal object data to the data processing unit ECU.

[0082] In the case of compressing sensor signal data (sampled values ​​of the sensor signal) describing the sensor signal, according to another variation, individual successive signal objects are progressively identified based on a feature vector signal that includes all temporally successive signal objects, hereinafter referred to as the total feature vector signal. For each identified signal object, a single feature vector signal is reconstructed, representing that identified signal object. This single feature vector signal is then transformed back into the signal object through a transformation in the form of inverse feature extraction, i.e., transformed into the time domain. Next, this (single) signal object is subtracted from the sensor signal, and feature extraction is then performed again on the remaining portion of the sensor signal so that it can be further processed as described above, more specifically, further processed using the aforementioned artificial neural network.

[0083] According to another embodiment of the invention, the present invention relates to a method for transmitting sensor signals from a transmitter to a data processing unit in a measurement system, particularly for distance measurement and especially for applications in vehicles, and particularly for transmitting ultrasonic sensor signals from an ultrasonic sensor to a data processing unit, wherein in this method...

[0084] - Provides sensor signals from sensors S1, S2, S3, ..., Sm.

[0085] - The sensor signal data describing the sensor signal is transmitted wirelessly or via wired connection from sensors S1, S2, S3, ..., Sm in a compressed manner to the data processing unit ECU.

[0086] - In order to compress the sensor signal data describing the sensor signal,

[0087] - By using feature extraction (FE) to extract signal waveform characteristics from the sensor signal,

[0088] - The total feature vector signal F1 is formed from the extracted signal waveform characteristics.

[0089] - Assign signal objects to the features of the total eigenvector signal F1 in the following way:

[0090] a) Using an artificial neural signal object recognition network NN0, signal objects are identified based on the features of the total feature vector signal F1.

[0091] b) For the identified signal object, another artificial neural single reconstruction network among multiple artificial neural single reconstruction networks NN1, NN2, ..., NNn is used to generate reconstructed single feature vector signals R1, R2, Rn-1, Rn.

[0092] c) Calculate the sum of all reconstructed individual eigenvector signals R1, R2, Rn-1, Rn, and use this sum as the total individual eigenvector signal RF.

[0093] d) The sum of individual eigenvector signals RF is inversely transformed into a signal object, more precisely, in particular, the sum of individual eigenvector signals is inversely transformed into a signal object by means of the inverse transform IFE of the feature extraction FE of the sensor signal.

[0094] e) Subtract the inversely transformed signal object from the sensor signal to form the residual sensor signal.

[0095] f) As long as the residual sensor signal is less than a pre-defined threshold signal, the assignment of signal objects to the features of the total feature vector signal F0 is terminated.

[0096] g) Otherwise, feature extraction (FE) is used to extract signal waveform characteristics from the residual sensor signal, and a total feature vector residual signal F0 is formed based on these signal waveform characteristics.

[0097] h) Based on the corresponding current total feature vector residual signal F0, repeat steps a) to g) until the residual sensor signal is less than a pre-given threshold signal, and

[0098] i) End the assignment of signal objects to the features of the total eigenvector signal F0, and

[0099] - For each identified signal object, generate signal object data representing that signal object, and

[0100] - Transmit these signal object data to the data processing unit ECU.

[0101] In another alternative to the data compression method according to the invention, the iterative process of extracting signal objects is implemented as follows: for each extracted signal object provided by the neural signal object recognition network, after reconstructing it into a single feature vector signal, the signal object, or more precisely, the time waveform of the sensor signal of the signal object, is formed from this single feature vector signal by transformation to the time domain. For this purpose, the corresponding single feature vector signal is inversely transformed into the time domain, more precisely, in particular by means of a transformation substantially inversely performed with respect to the previous feature extraction. The identified signal object, thus transformed into the time domain, is now subtracted from the sensor signal, resulting in a residual sensor signal, which is fed to the neural signal object recognition network after feature extraction. The neural signal object recognition network then identifies the next dominant signal object, and the above process continues iteratively. The process ends as soon as the residual sensor signal is less than a predetermined threshold signal.

[0102] In another variation of the data compression method according to the present invention, the above-mentioned inverse transformation is performed on a single feature vector signal formed for the identified signal object by means of multiple independent inverse feature extractions to perform the inverse transformation on these single feature vector signals.

[0103] All of the above variations of the data compression method are applicable: they can work not only with multiple neural single compression networks, but also with a single, unique neural single compression network.

[0104] The signal object data remaining after data compression includes a reduction in the amount of data used to identify the signal object category to which the signal object belongs, compared to the data used to describe the sensor signal and the data used to describe the identified signal object. This includes signal object parameter data, which describes the signal object to be assigned according to the signal object category used to describe variants different from the original signal object category, such as the overall shape of the signal object (e.g., triangular signal, extension, position within the sensor signal, slope, size, deformation). Using these parameters, the amount of data required to describe the signal object is significantly reduced.

[0105] Now, symbols are assigned to each signal object; these symbols, in this respect, represent the signal object data of the corresponding signal object. A symbol is the name of the signal object category to which the identified signal object is assigned. For example, there is a signal object category for triangular signal objects and another category for, for example, rectangular signal objects. This should be understood as illustrative only and should not be construed as a closed enumeration of signal object categories. The definition of signal object categories depends particularly on the application. The more signal object categories defined, the higher the data compression, but the more complex the signal processing associated with that data compression becomes.

[0106] By representing signal objects in the form of symbols (signal object data), the amount of data that still needs to be transmitted is significantly reduced. As mentioned above, the first data compression stage precedes this second stage of data compression, in which the signal waveform characteristics have been extracted based on the waveform of the sensor signal, that is, based on the sequence of sampled values.

[0107] In another advantageous embodiment of the invention, decompression can now, in principle, be performed in the data processing unit in reverse to the process described above. According to a variant also considered an independent variation of the invention, the process involves a method for decompressing compressed sensor signal data describing sensor signals of sensors S1, S2, S3, ..., Sn, particularly sensors used in vehicles, and particularly compressed sensor signal data describing ultrasonic sensor signals, wherein in this method...

[0108] - Provides compressed sensor signal data describing sensor signals, wherein these sensor signal data represent signal objects, which are assigned signal waveform characteristics of the sensor signals. These signal waveform characteristics are extracted from the sensor signals by means of feature extraction FE and form the features of the total feature vector signal F1.

[0109] For each signal object, the data describing the corresponding signal object is fed to another artificial neural single decompression network among multiple artificial neural single decompression networks ENN1, ENN2, ..., ENNn; or, a corresponding signal object is generated based on the data describing the signal object and the signal object is fed to another artificial neural single decompression network among multiple artificial neural single decompression networks ENN1, ENN2, ..., ENNn, where these neural single decompression networks ENN1, ENN2, ..., ENNn are parameterized, and these parameters are the same as those used by artificial neural networks when compressing sensor signal data.

[0110] - Each neural network, from its individual decompression network ENN1, ENN2, ..., ENNn, forms a reconstructed single feature vector signal ER1, ER2, ..., ERn, and...

[0111] - The reconstructed individual feature vector signals ER1, ER2, ..., ERn are summed to form a reconstructed total feature vector signal ER representing the total feature vector signal F1, which represents the decompression of the compressed sensor signal data.

[0112] Relatedly, the decompression of compressed sensor signal data describing sensor signals according to the present invention is based on compressing the sensor signal data according to the eigenvector signal of the sensor signal itself. A key feature of this decompression is the use of multiple artificial neural individual decompression networks, each of which is assigned to the signal object transmitted in compressed form from the sensor output signal. These individual neural decompression networks generate reconstructed individual eigenvector signals, which are summed to form a reconstructed total eigenvector signal. This total eigenvector signal represents the decompression of the compressed sensor signal data, i.e., ultimately represents the total eigenvector signal used in the formation of the compressed sensor signal data.

[0113] Regarding the aforementioned decompression of compressed sensor signal data, it is also possible to form reconstructed single feature vector signals ER1, ER2, ..., ERn through spatial or temporal multiplexing.

[0114] As described above, the variant of the present invention relating to the decompression of previously compressed sensor signal data is suitable for use in conjunction with the previously performed method described above for transmitting compressed sensor signal data.

[0115] As previously explained, according to the present invention, physical obstacle objects in the vehicle's surrounding environment are identified based on sensor signals from multiple sensors. Then, based on this identification, an environmental map of the vehicle can be generated, illustrating the distance of each obstacle from the vehicle, the location of these obstacles, particularly their orientation, type, nature, etc. For the purposes of this invention, signal notification within the vehicle is also understood as such an environmental map, wherein a minimum distance to an obstacle object is indicated optically and / or acoustically and / or tactilely (steering wheel vibration).

[0116] Regarding the identification of the presence of obstacle objects in the vehicle's environment, the present invention, according to another alternative, relates to a method for identifying the presence of obstacle objects in a detection area adjacent to the vehicle, particularly in the environment surrounding the vehicle, based on sensor signals provided by multiple sensors, especially multiple ultrasonic sensors, wherein in this method...

[0117] A) For each sensor, signal waveform characteristics are extracted by feature extraction from the sensor signal of that sensor, and feature vector signals EF1, EF2, EF3, ..., EFm representing the sensor signal are formed from these features.

[0118] B) The feature vector signals EF1, EF2, EF3, ..., EFm are fed as input signals to the artificial neural obstacle object recognition network ANN0. This network identifies at least one obstacle object based on the features of the feature vector signals EF1, EF2, EF3, ..., EFm, and stores the description of each obstacle object in separate obstacle object memories EO1, EO2, ..., EOp.

[0119] C) Input the information of each obstacle object memory EO1, EO2, ..., EOp as input data into the artificial neural single obstacle object recognition network RNN1, RNN2, ..., RNNp assigned to that obstacle object memory.

[0120] D) Each neural obstacle object recognition network RNN1, RNN2, ..., RNNp outputs obstacle object feature vector signals RO1, RO2, ..., ROp representing the obstacle objects in the associated obstacle object memories EO1, EO2, ..., EOp, and an intermediate feature vector signal (ROij, where i = 1, 2, ..., p, where p equals the number of obstacle objects, and j = 1, 2, ..., m, where m equals the number of sensors) with the same number of sensors. Each of these intermediate feature vector signals is assigned a unique sensor S1, S2, S3, ..., Sm.

[0121] E) For each sensor S1, S2, S3, ..., Sm, the intermediate feature vector signals (ROij, where i = 1, 2, ..., p, where p equals the number of obstacle objects, and j = 1, 2, ..., m, where m equals the number of sensors) output by the neural single obstacle object recognition networks RNN1, RNN2, ..., RNNp are summed to form the corrected feature vector signals EV1, EV2, EV3, ..., EVm.

[0122] F) For each sensor S1, S2, S3, ..., Sm, form residual feature vector signals EC1, EC2, EC3, ..., ECm by subtracting the corrected feature vector signals EV1, EV2, EV3, ..., EVm from the feature vector signals RC1, RC2, RC3, ..., RCm.

[0123] G) As long as the residual eigenvector signals EC1, EC2, EC3, ..., ECm are greater than the threshold signal, repeat steps B) to F) using the corresponding updated residual eigenvector signals EC1, EC2, EC3, ..., ECm, and...

[0124] H) Otherwise, the obstacle object feature vector signals RO1, RO2, ..., ROp output by the neural subnetworks RNN1, RNN2, ..., RNNp represent obstacle objects respectively, and potential obstacle objects within the detection area are determined based on the obstacle object feature vector signals RO1, RO2, ..., ROp, especially regarding their position, orientation and / or orientation and / or their type, nature and / or distance from the vehicle.

[0125] The method begins with the existence of feature vector signals representing sensor signals, which in turn originate from a specific sensor among multiple sensors used to detect the vehicle's environment. All these feature vector signals are fed into an artificial neural network for overall obstacle object recognition, which identifies one or more obstacle objects based on these features. The descriptive information of these one or more obstacle objects is then stored in separate obstacle object memories. An obstacle object memory exists for each obstacle object. Next, based on these obstacle object memories, the obstacle object feature vector signals are reconstructed using individual artificial neural networks for each memory. This is first achieved by generating intermediate feature vector signals, more specifically, for each obstacle object by using the feature vector signals assigned to that obstacle object, where all these signals are added and subtracted from the feature vector signals for the corresponding obstacle object to form a residual feature vector signal. This iterative process is repeated until the residual feature vector signal generated after each step is sufficiently small, i.e., less than a pre-defined threshold signal. Thus, information about the actual obstacle objects present in the vehicle's environment is stored in these obstacle object memories. Next, parameters of these obstacle objects can be displayed as required, such as location, orientation, distance from the vehicle, type, nature, etc.

[0126] Another aspect of the invention relates to the prediction of changes in identified, real-world obstacle objects within a prediction period and up to the point where updated parameters of the obstacle objects, which are actually present in the vehicle's environment, are presented. "Changes in obstacle objects" are understood as changes in their position and / or orientation and / or distance from the vehicle. In the prediction phase, prediction is performed by identifying "virtual" obstacle objects. That is, "virtual" obstacle objects describe changes in real-world obstacle objects within the prediction period. Ideally, at the end of the prediction period, the virtual obstacle objects will be substantially consistent with data regarding real-world obstacle objects, provided by the vehicle's (e.g., ultrasonic) measurement system at the start of the next prediction period. For this process, artificial neural networks are then used, which are correspondingly trained, and for this purpose, refer again to their more aforementioned definition, which should be understood as "artificial neural networks" within the framework of this invention.

[0127] According to another variation of the invention, the invention relates to a method for predicting potential changes in the distance between an obstacle object (particularly regarding its position, orientation, and / or orientation and / or its type, nature, and / or distance from the vehicle to the ranging system) within a detection area adjacent to a vehicle over a prediction time period, and particularly for predicting changes in the distance between the obstacle object and the ranging system due to the relative motion between the obstacle object and the ranging system, wherein in this method...

[0128] I. The prediction time period EP is divided into multiple successive prediction periods PP1, PP2, ..., PPq, and each prediction period PP1, PP2, ..., PPq is further divided into successive prediction cycles PZ1, PZ2, ..., PZp, the same number as the number of obstacle objects in the detection area. For each prediction period PP1, PP2, ..., PPq, the change of the obstacle object relative to the ranging system is predicted relative to the changes predicted in the previous prediction periods PP1, PP2, ..., PPq.

[0129] II. Before the start of each prediction time period EP and / or at the start of each prediction time period and / or with the start of the prediction time period, and thereby for the first prediction period PP1 for each obstacle object, obstacle object feature vector signals RO1, RO2, ..., ROp are provided. These obstacle object feature vector signals represent information about the obstacle object at points in time, such as the obstacle object's position, especially its orientation and / or orientation and / or distance relative to the ranging system. At these points in time, the obstacle object feature vector signals RO1, RO2, ..., ROp are determined based on current measurements taken by at least one sensor S1, S2, S3, ..., Sm, particularly at least one ultrasonic sensor, that detects the detection area regarding the potential presence of an obstacle.

[0130] III. Input the obstacle object feature vector signals RO1, RO2, ..., ROp into the artificial neural network DANN0.

[0131] IV. The neural network DANN0 generates prediction information VEO1, VEO2, ..., VEOp for each obstacle object, representing its position within prediction periods PZ1, PZ2, ..., PZp during the current prediction period PP1, PP2, ..., PPq.

[0132] V. Store the predicted information VEO1, VEO2, ..., VEOp for each obstacle object in the prediction memory VM1, VM2, ..., VMp.

[0133] VI. For each obstacle object, the information currently stored in the prediction memory VM1, VM2, ..., VMp is fed to another artificial neural single prediction network in multiple artificial neural single prediction networks MNN1, MNN2, ..., MNNp.

[0134] VII. Each individual neural prediction network MNN1, MNN2, ..., MNNp generates intermediate prediction feature vector signals RV11, RV12, ..., RVp1, which represent the predicted changes of the associated obstacle objects HO1, HO2, ..., HOp during the prediction period PZ1, PZ2, ..., PZp.

[0135] VIII. For each obstacle object HO1, HO2, ..., HOp, modify the obstacle object feature vector signals RO1, RO2, ..., ROp using the intermediate predicted feature vectors RV11, RV12, ..., RVp1.

[0136] IX. The altered obstacle object feature vector signals RO1, RO2, ..., ROp are then fed into the overall prediction network DANN0.

[0137] X. For each prediction period PP1, PP2, ..., PPq, steps III to IX are performed simultaneously for all obstacle objects HO1, HO2, ..., HOp, or sequentially, more precisely, for each prediction period PZ1, PZ2, ..., PZp, these steps are performed for another obstacle object HO1, HO2, ..., HOp, wherein the order in which changes to obstacle objects HO1, HO2, ..., HOp are checked remains the same accordingly.

[0138] XI. At the latest at the end of the prediction time period EP, the contents of the prediction memory VM1, VM2, ..., VMp are fed to the artificial neural reality simulation network MNN0, which outputs the reality simulation feature vector signal VRV, which represents the predicted current changes of obstacle objects HO1, HO2, ..., HOp.

[0139] The method according to the present invention is based on the fact that ranging systems have dead time or pause time between measurements. In the case of ultrasonic ranging systems, this means that a certain amount of time must pass between emitting an ultrasonic signal and receiving a reflected signal. Evaluating the received ultrasonic signal to determine potential obstacles in the detection area of ​​the measuring system also requires time. It is desirable to obtain information about how the positions of the one or more obstacles have changed during this period, from the point in time when one or more locations of one or more obstacles were detected based on the previous measurement until the point in time when an updated measurement (i.e., a measurement for the next measurement period) is available.

[0140] This is achieved using multiple artificial neural networks that are trained accordingly. Information about the current positions of obstacle objects is fed into a central neural prediction network. The predicted changes in the positions of the obstacles as the prediction period progresses (which essentially corresponds to the current measurement period being performed in parallel) are also fed into this central prediction network. The central neural prediction network generates prediction information for each obstacle object, which is stored in a corresponding prediction memory. Based on this prediction information, intermediate prediction feature vector signals describing this change are generated in individual neural prediction networks, an number equal to the number of obstacle objects. These changes are then fed into the central neural prediction network for the next prediction cycle within one of the multiple prediction periods of the prediction time. The central prediction network simulates possible changes for each obstacle object based on the information it has acquired during the training phase. The contents of the prediction memory can be retrieved at any point in this process to output a general vector representation of the predicted reality of the obstacle object arrangement after being input into the neural reality simulation network. Information about the predicted "reality" is typically generated or retrieved no later than the end of the prediction time period. The prediction time period comprises q prediction periods, each of which is further divided into multiple prediction cycles, the number of which equals the number of identified obstacle objects. Predicting changes in obstacle objects can be achieved not only through temporal multiplexing but also through spatial multiplexing.

[0141] In the foregoing, the invention has been described in particular in relation to its application in vehicles. Generally, the invention relates to ranging systems, and especially to such systems that operate based on ultrasound. However, radar-based measuring systems can also operate with respect to data processing as specified in the invention. Optical measuring systems can also be used.

[0142] According to the present invention, the above-mentioned task can also be solved alternatively by means of a method for operating an ultrasonic sensor, the method comprising the following steps:

[0143] - Detecting ultrasonic received signals;

[0144] - Provides a signal-free ultrasonic echo signal model 610;

[0145] - Perform the following steps at least once, or optionally repeatedly:

[0146] - Subtract the reconstructed ultrasound echo signal model 610 from the ultrasound received signal 1 to form the residual signal 660.

[0147] - Execute a method for identifying signal objects in residual signal 660.

[0148] - Give the ultrasonic echo signal model 610 ( Figure 15 (b to 16h) Supplement the signal waveforms of the identified signal objects 600 to 605;

[0149] - If the absolute value of the residual signal is lower than the absolute value of a pre-given threshold signal, then the repetition of these steps ends;

[0150] - Transmit symbols for at least a portion of the identified signal object and optionally use this information.

[0151] Furthermore, in order to solve the above-mentioned problems, an ultrasonic sensor system is proposed according to the present invention, which has the following features:

[0152] - First ultrasonic sensor; and

[0153] - At least one second ultrasonic sensor; and

[0154] - Computer system,

[0155] - Each of the at least two ultrasonic sensors implements a method for transmitting sensor data from the respective ultrasonic sensor to a computer system, the method comprising the following steps:

[0156] - It emits an ultrasonic pulse train α, and

[0157] - Receives ultrasonic signals and forms ultrasonic received signal β, and

[0158] - Data compression of the ultrasonic received signal is performed using the reconstructed ultrasonic echo signal model 610 to generate compressed data γ, and

[0159] - Transmit compressed data to the computer system δ.

[0160] - Within the computer system, at least two compressed ultrasonic received signals are decompressed or reconstructed into reconstructed ultrasonic received signals using a reconstructed ultrasonic echo signal model, and

[0161] - The computer system uses these reconstructed ultrasonic received signals to perform object recognition of objects in the environment of the ultrasonic sensor.

[0162] In an advantageous extension of this variant of the invention, the computer system of the ultrasonic measurement system can perform object recognition of objects in the sensor's environment by means of reconstructed ultrasonic received signals and additional signals from other sensors, especially radar sensors, and / or the computer system can create an environmental map of the sensor or part of the device that is the sensor based on the identified objects.

[0163] The artificial neural networks mentioned within the framework of this invention are correspondingly trained, more precisely, trained using training data that is known in principle for the learning phase of the neural network. Within the framework of this invention, artificial neural networks are any type of artificial intelligence concept. According to this invention, artificial neural networks are used to generate data with different contents and signals formed therefrom. Artificial neural networks are known to be based on data analysis using relatively large amounts of data, which are fed said relatively large amounts of data during the training and learning phases, and said relatively large amounts of data are obtained through experimental testing or simulation before being used to analyze currently existing input signals (so-called machine learning or deep learning). All types of known and future single-layer or multi-layer artificial neural networks can be used, as well as artificial neural networks that learn and / or build themselves during real-world operation. For the purposes of this invention, the term "artificial neural network" includes all known and future data / signal analysis and data / signal processing tools that enable what is understood in a narrow and broad sense as "artificial intelligence" and / or will continue to be understood as "artificial intelligence" in the future.

[0164] The artificial neural network used according to the present invention is fed with data obtained experimentally or otherwise through simulation during the training or learning phase. For example, data is supplied to the signal object recognition network NN0, describing the assignment of signal waveform characteristics or groups of continuous signal waveform characteristics to their respective signal objects based on feature vectors or feature vector signals. Data is supplied to individual reconstruction networks NN1, NN2, ..., NNn, describing the assignment of signal objects to feature vectors or feature vector signals representing these signal objects. Data is fed to individual decompression networks ENN1, ENN2, ..., ENNn, describing the assignment of signal objects to feature vectors or feature vector signals representing these signal objects. The overall obstacle object recognition network ANN0 includes data including the assignment of obstacle objects to signal waveform characteristics of multiple sensor signals and describing the corresponding feature vectors or feature vector signals. Data is supplied to individual obstacle object recognition networks RNN1, RNN2, ..., RNNp, describing the assignment of obstacle objects to feature vectors or feature vector signals, which in turn represent the signal waveform characteristics of multiple sensor signals.

[0165] The overall prediction network DANN0 contains data including the assignment of signal waveform characteristics from obstacle objects to multiple sensor signals and descriptions of the associated feature vectors or feature vector signals. Individual prediction networks MNN1, MNN2, ..., MNNp are supplied with data describing the assignment of obstacle objects to a total feature vector or total feature vector signal in an environment such as a vehicle, representing all obstacle objects in that environment.

[0166] Various details and measures are described below, which can be used in conjunction with the implementation of the method according to the invention. The invention has been described above or subsequently primarily based on its use in a vehicle, where a measurement system is used to check the vehicle's environment for the presence of obstructive objects. However, in general, the invention relates to a measurement system for determining the distance of the system from various objects, wherein the types of signals emitted and received can be diverse.

[0167] This invention is based on creating feature vectors and feature vector signals by feature extraction from sensor signals.

[0168] The following describes an example of creating a feature vector signal. A feature vector signal is a sequence of feature vectors. A "feature," such as a characteristic of the waveform of a signal provided by a sensor over time, can include its value over time (signal level), the value of the first (and / or higher-order) derivative of the signal waveform over time, the value of the integral of the signal waveform over time, the logarithm of the signal waveform over time, and so on. That is, the basis for creating feature vectors is various curves formed, for example, from the waveform of a sensor signal over time through mathematical operations. If these curves are now sampled, there are multiple values ​​for each sampling time point, and these values ​​form a feature vector at that specific time point. This sequence of feature vectors then forms a feature vector signal. Each feature vector has multiple values ​​(also called parameters). The sequence of these parameters, each assigned to the same curve, then forms a parameter signal.

[0169] In other words, feature vector extraction can also be described as follows:

[0170] The signal to be processed is handled through techniques such as differentiation, integration, filtering, transformation, delay, and threshold comparison. This typically increases the dimensionality of the signal dramatically from 1, for example, to 24 or more. What exactly is done is irrelevant here. Feature extraction is highly application-specific.

[0171] - The individual dimensions of the resulting eigenvectors generally do not have optimal significance with respect to the commonly used test dataset. Optimal significance exists if, within the allowed parameter range for a single dimension of the eigenvector, 50% of the eigenvectors are above a threshold in the middle of that range, and 50% are below that threshold. To achieve maximum significance, the dimensions of the eigenvectors are typically reduced by multiplying with the LDA matrix, and the eigenvectors are deformed or rotated to establish this significance.

[0172] An eigenvector can be understood as a multidimensional signal stream of sampled values ​​being processed, and for example, a multidimensional sampled value with multiple signals, more precisely, multidimensional sampled values ​​at corresponding sampling points in time. A sequence of eigenvectors can be called an eigenvector signal. This signal has multiple dimensions. The signal corresponding to the unique specific dimension of this multidimensional eigenvector signal is the parameter signal assigned to that dimension.

[0173] Subsequently, the above process, and especially the correspondence between signal waveform characteristics, signal objects, and features therein or in the eigenvector signal, should be described again with reference to specific examples.

[0174] An increasing signal level, starting from a minimum value less than a threshold and increasing to a maximum value greater than that threshold within a specific time interval, should be considered an example of a first signal waveform characteristic. It should also be assumed that this first signal waveform characteristic is followed by a second signal waveform characteristic, i.e., a decreasing signal level waveform. That is, the resulting signal object would be a triangular signal with a specific slope and extension of its edges and a specific "centroid" (the orientation of the maximum value on the time axis). Then, the sequence formed by the sampling time points of this triangular signal and its curve waveform obtained through, for example, mathematical operations (see the example of such mathematical operations described above) represents the parameters (or values) of the eigenvectors of the characteristic vector signal representing the triangular signal. That is, based on the eigenvector signal, the signal object can be assigned to a "triangular signal," where parameters further describing the triangular signal, such as the slope / inclination of the edges, the orientation and magnitude of the maximum value, etc., can also be determined. In this respect, the eigenvector signal describes the signal object of the sensor signal, the "triangular signal." That is, when transmitting sensor signal data, it is now only necessary to transmit the symbol for the signal object category "triangle signal" and a few other parameters to describe the specific design of the triangle signal in more detail. This results in significant data compression and the advantages of the overall obstacle object recognition system already described above.

[0175] As described above, the technical teachings from the prior art all derive from the idea that the identification of obstacles in the vehicle's environment is performed in the ultrasonic sensor, and the object data is transmitted only after the obstacle object is identified. However, since the synergistic effect of using multiple ultrasonic transmitters is lost in this case, it has been recognized within the framework of the present invention that it is not reasonable to transmit only the echo data of the ultrasonic sensor itself, but rather to transmit the data of all ultrasonic sensors and evaluate the data of multiple sensors only in the central computer system (data processing unit). However, for this purpose, the data compression for transmission via a data bus with a smaller bus bandwidth must be performed differently than in the prior art. In this way, synergistic effects can then be developed. Thus, for example, it is conceivable that the vehicle has more than one ultrasonic sensor. In order to distinguish between the two sensors, it is reasonable that the two sensors transmit with different codes. However, unlike the prior art, both sensors should now detect the ultrasonic echoes radiated by the two ultrasonic sensors and transmit these ultrasonic echoes to the central computer system after appropriate compression, where the ultrasonic received signal is reconstructed. Object identification is performed only after this reconstruction (decompression). This also enables the fusion of ultrasonic sensor data with other sensor systems such as radar.

[0176] A method for transmitting sensor data from a sensor to a computer system is proposed. This method is particularly suitable for transmitting data of ultrasonic received signals from an ultrasonic sensor to a controller of a computer system, such as a vehicle's ranging system. Based on... Figure 1 This will be used to explain the method.

[0177] According to one embodiment of the method, an ultrasonic pulse train is first generated and then transmitted into free space, typically into the environment of a vehicle. Figure 1 Step α). Here, the ultrasonic pulse train consists of multiple acoustic pulses that occur sequentially to each other at ultrasonic frequencies. This ultrasonic pulse train is formed by a mechanical oscillator in the ultrasonic transmitter or ultrasonic transducer slowly initiating and then decaying through oscillation. Then, the ultrasonic pulse train emitted by this exemplary ultrasonic transducer is reflected at an object in the vehicle's environment and received as an ultrasonic signal by an ultrasonic receiver or the ultrasonic transducer itself, and converted into an electrical received signal (…). Figure 1 Step β). Particularly preferably, the ultrasonic transmitter is the same as the ultrasonic receiver and is hereinafter referred to as the transducer. However, the principles described below can also be applied to receivers and transmitters that are arranged and / or constructed separately. A signal processing unit is present in the proposed ultrasonic sensor, which now analyzes and compresses the thus received electrical signal (…). Figure 1 Step γ) is used to minimize the required data transmission and provide free space for the controller to send status reports and other control commands to the signal processing unit or ultrasonic sensor system. The compressed electrical received signal is then transmitted to the computer system. Figure 1 Step δ).

[0178] Therefore, the method according to the above-described variant and the advantageous design scheme described below is used to transmit sensor data, especially ultrasonic sensor data, from the sensor to a computer system, particularly a computer system in a vehicle. The method begins by emitting a train of ultrasonic pulses ( Figure 1 Step α). Then, the ultrasonic signal is received and an electrical receiving signal is generated ( Figure 1 Step β) and performing data compression on the received signal ( Figure 1 Step γ) to generate compressed data ( Figure 1 The step γ) involves detecting at least two or three or more predetermined characteristics. Preferably, the electrical received signal is obtained by sampling ( Figure 2 Step γa) is converted into a sampled received signal, which consists of a stream of time-discrete sampled values. Here, each sampled value can typically be assigned a sampling time point as a timestamp. This can be achieved, for example, through wavelet transform (…). Figure 2Compression is achieved through step γb). To this end, the received ultrasound signal, in the form of a sampled received signal, can be compared with, for example, predetermined signal basic shapes stored in a library, by calculating the correlation integral between the predetermined signal basic shape and the sampled received signal (the definition of this term is, for example, in Wikipedia). Hereinafter, the predetermined signal basic shape is also referred to as a signal object category. By calculating this correlation integral, the spectral values ​​belonging to each of these prototype signal object categories are determined. Since this is done continuously, these spectral values ​​are themselves time-discrete instantaneous spectral values, where a timestamp can be reassigned to each spectral value. An alternative, but mathematically equivalent, approach is to use an optimal filter (matched filter) for each predetermined signal object category (signal basic shape). Since multiple prototype signal object categories are typically used, and these categories may also undergo different time spreads, this typically results in a time-discrete multidimensional vector stream of different corresponding time spreads and their spectral values ​​for different prototype signal object categories, wherein a timestamp is reassigned to each of these multidimensional vectors. Each of these multidimensional vectors is a so-called feature vector. Therefore, a time-discrete feature vector stream is involved. Preferably, a timestamp is reassigned to each of these feature vectors (…). Figure 2 Step γb).

[0179] Therefore, the time dimension is also obtained through continuous time offsets. In this way, the eigenvectors of the spectral values ​​can also be supplemented with past values ​​or values ​​dependent on those eigenvectors, such as time integrals, derivatives, or filtered values ​​of one or more of these values. This can further increase the dimensionality of these eigenvectors within the eigenvector data stream. Therefore, to keep the following costs low, it is reasonable to limit the extraction of eigenvectors from the sampled input signal of the ultrasonic sensor to a small number of prototype signal object categories. Thus, an optimal filter (matched filter) can then be used, for example, to continuously monitor the presence of these prototype signal object categories in the received signal.

[0180] Here, isosceles triangles and bimodal peaks can be exemplarily referred to as particularly simple prototype signal object categories. These prototype signal object categories typically consist of a pre-defined spectral coefficient vector, i.e., pre-defined prototype eigenvector values.

[0181] To determine the importance of the spectral coefficients of the eigenvectors of an ultrasonic echo signal, the numerical distance (Betrag) of the elements of these characteristics, i.e., the vector of the current spectral coefficients (eigenvectors), is determined to be at least one prototype combination of these characteristics (prototypes) in the form of a prototype signal object category, which is symbolically represented by a pre-given prototype eigenvector (prototype or prototype vector) from a pre-given prototype signal object category vector library. Figure 2 Step γd). Preferably, the spectral coefficients of the eigenvectors are normalized before being correlated with the prototype ( Figure 2 Step γc). The distance determined in this distance determination can, for example, consist of the sum of all differences between a spectral coefficient of a pre-given prototype feature vector (prototype or prototype vector) of the corresponding prototype and the corresponding normalized spectral coefficient of the current feature vector of the ultrasonic echo signal. The root of the sum of the squares of all differences between a pre-given prototype feature vector (prototype or prototype vector) of the prototype and the corresponding normalized spectral coefficient of the current feature vector of the ultrasonic echo signal forms the Euclidean distance. However, this distance is usually too complex to obtain. Other methods for obtaining this distance are conceivable. Next, each pre-given prototype feature vector (prototype or prototype vector) can be assigned a sign and, if necessary, parameters, such as the distance value and / or amplitude before normalization. If the distance thus determined is below a first threshold and is the minimum distance between the current feature vector value and one of the pre-given prototype feature vector values ​​(prototype or prototype vector values), then the sign of the pre-given prototype feature vector value is again used as the identified prototype. Thus, the identified prototype and timestamp pairing of the current feature vector are formed. Next, preferably, data is transmitted to the computer system only when the distance is below a first threshold and the identified prototype is the prototype to be transmitted. Figure 2 Step δ), here is the symbol that best represents the identified prototype, along with, for example, the distance and the time of occurrence (timestamp). That is, it can also store unrecognizable prototypes, such as those used for noise, i.e., those without reflection, etc. This data is not important for obstacle recognition and therefore should not be transmitted if necessary. That is, if the determined distance between the current feature vector value and the pre-given prototype feature vector value (the prototype or the value of the prototype vector) is lower than the first threshold, then the prototype is identified ( Figure 2 Step γe). That is, instead of transmitting the ultrasound echo signal itself, only the symbol sequence of typical time signal waveforms identified within a specific time period and the timestamps belonging to these signal waveforms are transmitted. Figure 2Step δ). Therefore, preferably, for each identified signal object, only the symbol for the identified signal shape prototype, its parameters (e.g., the amplitude and / or time stretch of the envelope of the (ultrasonic echo) signal), and the reference time point (timestamp) at which the signal shape prototype appears are transmitted as the identified signal object. This eliminates the need to transmit time points where individual sample values ​​or thresholds are exceeded by the envelope of the sampled received signal (ultrasonic echo), etc. In this way, this selection of important prototypes results in significant data compression and a reduction in bus bandwidth required for fast transmission of large amounts of data in other cases.

[0182] That is, the presence of a combination of characteristics is quantitatively detected after forming an estimate—for example, the inverse distance between representations of prototype signal object categories in the form of a pre-given prototype feature vector (prototype or prototype vector)—and if the value of the estimate (e.g., the inverse distance) is higher than a second threshold or the inverse estimate is lower than a first threshold, then compressed data is transmitted to the computer system. Therefore, the signal processing unit of the ultrasonic sensor performs data compression of the received signal to produce compressed data.

[0183] To make it clearer, the distance determination (in conjunction with, for example, classifiers) should be reiterated here.

[0184] This distance determination is also known in classification from statistical signals. Here, as examples of classifiers, logistic regression, cuboid classifiers, distance classifiers, nearest neighbor classifiers, multinomial classifiers, clustering methods, artificial neural networks, and latent class analysis should be mentioned.

[0185] exist Figure 7 An example of a classifier is illustrated in the diagram.

[0186] The physical interface 101, for example, controls the ultrasound transducer 100 and causes the ultrasound transducer to emit, for example, ultrasound transmission pulses or ultrasound transmission pulse trains. The ultrasound transducer 100 receives signals not present in... Figure 7The ultrasonic pulses or bursts reflected by the obstacle object are drawn in the image. These ultrasonic pulses or bursts have amplitude variations, delays, and are generally deformed due to the properties of the obstacle object reflecting them. Furthermore, multiple obstacle objects are typically present in a vehicle environment, which contribute to the modulation of the reflected ultrasonic waves. An ultrasonic transducer 100 converts the received reflected ultrasonic waves into an ultrasonic transducer signal 102. The physical interface converts this ultrasonic transducer signal into an ultrasonic echo signal 1, typically through filtering and / or amplification. A feature vector extractor 111 extracts signal waveform characteristics (features) from the ultrasonic echo signal 1. Preferably, the physical interface 101 transmits the ultrasonic echo signal 1 to the feature vector extractor 111 as a time-discrete received signal consisting of a sequence of sampled values. Here, it is preferable to assign a time date (timestamp) to each sampled value. Figure 7 In the example, the feature vector extractor 111 has m (m is a positive integer) optimal filters (optimal filter 1 to optimal filter m). These optimal filters are used to determine intermediate parameter signals 123 from the sampled value sequence of the ultrasonic echo signal 1, preferably by means of appropriate filters (e.g., optimal filters), each intermediate parameter signal relating to the existence of a signal basic object preferably assigned to the corresponding intermediate parameter signal. The resulting intermediate parameter signals 123 are also designed as a time-discrete sequence of corresponding intermediate parameter signal values, which are preferably associated with dates (time stamps). Therefore, it is preferable to assign exactly one time date (time stamp) to each intermediate parameter signal value.

[0187] Subsequent significance enhancer 125 performs a matrix multiplication of the vector of the intermediate parameter signal values of the intermediate parameter signal 123 with a so-called LDA matrix 126. This LDA matrix is typically determined at the construction time point by means of statistical signal processing and pattern recognition methods. The significance enhancer generates the feature vector signal 138 in this way. In this case, the significance enhancer maps the m intermediate parameter signal values to n (n is an integer) parameter signal values of the feature vector signal 138. That is, the feature vector signal 138 typically includes n parameter signals. Preferably n < m. It should be noted in this regard that in statistical signal theory and in pattern recognition, the term "Feature-Vektor" is often also referred to as a feature vector (also denoted as "feature vector" subsequently). Thus, the feature vector signal 138 is designed in particular as a time-discrete sequence of feature vector signal values, each of which has the n parameter signal values of the preferably n parameter signals of the feature vector signal 138, and these feature vector signal values include these parameter signal values as well as other parameter signal values having the same time date (timestamp) respectively. In this case, n is the dimension of each feature vector signal value, and these feature vector signal values are preferably the same from one feature vector value to the next. In this regard, the feature vector signal values are vectors with timestamps, and the vector includes a plurality of, preferably n, parameter signal values. Each such formed feature vector signal value is assigned the corresponding time date (timestamp).

[0188] Now, then, the time waveform of the feature vector signal 138 in the resulting n-dimensional phase space is evaluated and the recognized signal basic object is inferred in the case of determining an evaluation value (such as a distance).

[0189] For this purpose, the distance determiner (or classifier) 112 compares the current feature vector signal value of the feature vector signal 138 with a plurality of prototype feature vector signal values previously stored in the prototype database 115. This will be elaborated in more detail below. Here, the distance determiner (or classifier) 112 determines an evaluation value for each feature vector signal value prototype in the inspected feature vector signal value prototypes of the prototype database 115, and this evaluation value indicates how much the corresponding feature vector signal value prototype in the prototype database 115 is the same as the current feature vector signal value. This evaluation value is referred to as a distance below. Preferably, these feature vector signal value prototypes are each preferably assigned to exactly one signal basic object. Then, the feature vector signal value prototype in the prototype database 115 having the smallest distance from the current feature vector signal value is the most similar to the current feature vector signal value. If the distance is less than a pre-given threshold, this feature vector signal value prototype in the prototype database 115 represents the signal basic object 121 recognized with the highest probability.

[0190] The identification process is performed repeatedly so that the determined signal basic object sequence is obtained from the time series of the identified signal basic object 121.

[0191] Next, a possible signal object 122 is determined by identifying the sequence in the pre-given sequence of basic signal objects in the signal object database 116 that is most similar to the determined sequence of basic signal objects. As described above, the signal object here consists of a time series of basic signal objects. Here, symbols are typically predefined and assigned to the signal objects in the signal object database 116.

[0192] For example, this estimation of the signal basic object sequence can be achieved using a Viterbi estimator 113. In the simplest case, the number of signal basic objects identified within a pre-given time period whose positions in the identified signal basic object sequence match the positions of the expected signal basic objects in the pre-given signal basic object sequence in the signal object database 116, minus the number of signal basic objects identified within the pre-given time period whose positions in the identified signal basic object sequence do not match the positions of the expected signal basic objects in the pre-given signal basic object sequence in the signal object database 116, is used as the evaluation value for consistency. In this way, the Viterbi estimator determines an evaluation value for each signal object in the pre-given signal objects of the signal object database 116, where the signal objects of the signal object database 116 consist of a pre-given sequence of expected signal basic objects and are associated with corresponding symbols.

[0193] In other words, it is checked whether the point pointed to by the n-dimensional feature vector signal 138 in the n-dimensional phase space is closer than a predetermined point in the n-dimensional phase space by a predetermined maximum distance in its path traversing the n-dimensional phase space in a predetermined time order. That is, the feature vector signal 138 has a time waveform. Next, an evaluation value (e.g., distance) is calculated, which may reflect the probability of the existence of a specific sequence. Then, preferably within the Viterbi estimator 113, the evaluation value reassigned with the time date (timestamp) is compared with a threshold vector to form a Boolean result, which may have a first value and a second value. If the Boolean result has a first value for the time date (timestamp), the symbol of the signal object and the time date (timestamp) to which the symbol was assigned are transmitted from the sensor to the computer system. Thus, the identified signal object 122 is preferably transmitted along with its parameters. If necessary, other parameters can be transmitted according to the identified signal object.

[0194] In this regard, to clarify further, the processing of the eigenvector signal 138 should be discussed again. Preferably, the ultrasonic echo signal 1 is a sequence of quantization vectors—ultrasonic echo signal values—but the components of these ultrasonic echo signal values, i.e., the measured parameters, are typically not entirely independent of each other. Each ultrasonic echo signal value of the ultrasonic echo signal 1 itself typically has too low selectivity for precise identification of signal objects in complex relationships. This is precisely the shortcoming of the prior art. Through the physical interface 101 (see...) Figure 7 Typically, one or more such quantization vectors are created at regular or periodic intervals based on a continuous stream of analog physical values ​​of the ultrasonic echo signal 1, forming a multidimensional ultrasonic echo signal value data stream that is quantized in time and value in the form of the ultrasonic echo signal 1.

[0195] The resulting multidimensional ultrasonic echo signal 1, in the form of one or more quantized vector streams, is first divided into single frames of defined length in the first processing step, filtered, normalized, then orthogonalized, and appropriately deformed, if necessary, through nonlinear mapping—such as logarithmic transformation and cepstral analysis. This is achieved through… Figure 7 The optimal filter block in the feature vector extractor 111 is used to outline the intermediate parameter signal 123. That is, instead of the optimal filter, other signal processing structures can also be envisioned to generate the intermediate parameter signal 123. For example, the derivative of the ultrasonic echo signal value generated in this way can also be calculated. Finally, the determined intermediate parameter signal 123 is significantly enhanced relative to the actual feature vector signal 138 in the significance enhancement unit 125. As described, this can be achieved, for example, by multiplying a multidimensional quantization sector with a so-called pre-given LDA matrix 126.

[0196] The identification now performed in the distance determiner (or classifier) ​​112 can be performed in different ways, for example:

[0197] a) Through neural networks, or

[0198] b) Through HMM identifier

[0199] c) Through Petri network.

[0200] Here, the HMM recognizer is described again as an example. Figure 7 ):

[0201] With the aid of the aforementioned predefined LDA matrix 126, the intermediate parameter data stream 123 is thus mapped from the multidimensional input parameter space to a new parameter space by the significance enhancer 125, thereby maximizing its selectivity. In this case, the components of the newly obtained transformed eigenvector are selected not based on the true physical parameters or other parameters, but based on the maximum significance, which results in the aforementioned maximum selectivity.

[0202] LDA matrix 126 is typically computed offline beforehand at the construction time based on an example data stream of a dataset with known signal objects, i.e., a dataset obtained using a pre-given signal waveform structure, through a training step.

[0203] If it is ensured that all elements of the method performed by the distance determiner (or classifier) ​​112 implement at least locally invertible functions, the deviation of the signal waveform can be considered in the form of an approximately linear transformation function.

[0204] The prototypes in the example data stream of pre-given prototype signal waveforms (prototype signal basic objects) from coordinates in the new parameter space are calculated during the construction phase and stored in the prototype database 115 for later re-identification. In addition to these statistics, the prototype database may also contain instructions for a computer system, which should be implemented in response to the successful or unsuccessful identification of the corresponding signal basic object prototype. Typically, this computer system will be the computer system for the sensor system.

[0205] Therefore, the feature vector signal values ​​of the feature vector signal 138, which are output in the laboratory by the feature extractor 111 for the pre-given signal basic object of the ultrasonic echo signal 1, are protected in the prototype database 115 as the signal basic object prototype.

[0206] In later execution, the eigenvector signal value of the eigenvector signal 138 is now compared with these previously stored, i.e., taught, signal primitive object prototypes 115 by, for example, calculating the Euclidean distance between the quantized vector in the coordinates of the new parameter space and all these previously stored signal primitive object prototypes 115 in the distance determiner 112. In this case, at least two identifications are performed:

[0207] 1. Does the identified eigenvector value of eigenvector signal 138 correspond to, and with what probability and reliability, one of the pre-stored signal basic object prototypes in prototype database 115?

[0208] 2. If it is one of the signal basic object prototypes already stored in prototype database 115, then which signal basic object prototype is it and with what probability and reliability is it?

[0209] For initial identification, virtual prototypes are typically stored in a prototype database 115 of signal basic object prototypes. These signal basic object prototypes should cover as many parasitic parameter combinations as possible that occur during runtime. The aforementioned signal basic object prototypes are stored in prototype database 115.

[0210] For the signal basic object prototypes in prototype database 115, the distance can be determined at the build time for each pair of two different signal basic object prototypes in prototype database 115 according to the method applied in distance determiner 112. Then, the minimum prototype distance is obtained. This minimum prototype distance is preferably also halved to half the minimum prototype distance in prototype database 115 or in distance determiner.

[0211] If, for example, the distance determined by distance determiner 112 between the current eigenvector signal value of eigenvector signal 138 and the signal basic object prototype in prototype database 115 is less than half of the minimum prototype distance, then the signal basic object prototype is evaluated as identified. From this point onward, it can be ruled out that other distances to other signal basic object prototypes in prototype database 115 calculated during further continued searching may provide even smaller distances. The search can then be interrupted, which halves the average time and thus saves sensor resources.

[0212] Here, the calculation of the minimum Euclidean distance can be performed, for example, using the following formula:

[0213]

[0214] Here, dim_cnt represents the dimension index up to the maximum dimension dim of the feature vector 138.

[0215] FV dim_cnt The parameter value representing the eigenvector signal value corresponding to the index dim_cnt for eigenvector 138.

[0216] Cb_cnt represents the number of the basic signal object prototype in prototype database 115.

[0217] Correspondingly, Cb CB_cnt,dim_cnt The parameter value corresponding to dim_cnt represents the entry of the signal basic object prototype in the prototype database 115, which is assigned to the signal basic object prototype corresponding to Cb_cnt.

[0218] Dist FV_CbE This represents the minimum Euclidean distance obtained here as an example. In the case of searching for the minimum Euclidean distance, the number Cb_cnt that produces the minimum distance is recorded.

[0219] To illustrate, let's look at some examples of assembly code:

[0220] Code Start

[0221] Mov Cb_cnt, #Cb_anz Initialize prototype database vector counter

[0222] Mov C, #0 Initialize register C with 0.

[0223] Mov dist, maxvalue initializes the distance using the maximum value.

[0224] `Mov num, not_valid_num` initializes the number of the next neighbor with an invalid value.

[0225] Mov Cb_adr, Cb_badr initializes the prototype database address using the base address of the prototype database.

[0226] Label_A: / / next vector

[0227] Mov SP, #0 Initialize buffer memory

[0228] Mov dim_cnt, #dim initializes the dimension counter using the feature vector dimensions.

[0229] Label B: / / next dimension

[0230] MovA, $Cb_adr absolutely loads values ​​from the prototype database address.

[0231] SubA, $Fv_adr, dim_cnt subtracts values ​​from the eigenvector values ​​absolutely or relatively.

[0232] Mov BA uses the result to load the B register.

[0233] Multiply A and B by MulA B (= A) 2 )

[0234] AddA, SP adds the result to the intermediate results.

[0235] Mov SP, A and record

[0236] Dec dim_cnt Next vector component

[0237] Inc Cb_adr increments the prototype database pointer by one.

[0238] jnz dim_cnt, Label B But only when it is not the last one

[0239] Cmp SP, dist evaluates the prototype database entries (signal basic object prototypes).

[0240] jmpgt Label C

[0241] Mov dist, SP If there are better entries than the best value so far

[0242] Mov num, Cb_cnt records the entry number and distance.

[0243] Label C:

[0244] dec_Cb_cnt Next prototype database entry

[0245] jnz Cb_cnt, Label A But only when it is not the last one

[0246] End of code

[0247] The confidence level for correct identification is derived from the divergence of the basic data stream based on the prototype of the basic object of the signal and the distance of the current eigenvector value of the eigenvector signal 138 from its centroid.

[0248] Figure 8 Different recognition scenarios are illustrated. For simplicity, a representation for two-dimensional feature vector signals is chosen, where each feature vector signal value includes a first parameter value and a second parameter value. Here, this is simply for the purpose of better presenting the methodology on a two-dimensional sheet of paper. In reality, the feature vector signal values ​​of feature vector signal 138 are usually always multi-dimensional.

[0249] Draw the centroids of the different prototypes 141, 142, 143, and 144. As described above, half the minimum distance between these signal basic object prototypes in prototype database 115 can now be stored in prototype database 115. This will then be a global parameter that is equally valid for all signal basic object prototypes in prototype database 115. The decision is performed using this minimum distance. However, this decision is predicated on the fact that the scattering of the signal basic object prototypes, labeled by their centroid orientations 141, 142, 143, and 144, is more or less the same. This is also true if the distance determination (or other evaluation) performed by distance determiner 112 (or classifier) ​​is optimal. This corresponds to a circle around the centroids 141, 142, 143, and 144 of each signal basic object prototype in prototype database 115.

[0250] However, this is rarely achievable in practice. Therefore, if the spread widths of the signal basic object prototypes in the prototype database 115 are stored separately, an improvement in recognition performance can be achieved. This corresponds to a circle with a radius specific to each signal basic object prototype in the prototype database 115. The disadvantage is the increased computational power.

[0251] If the scattering width of the signal basic object prototypes in the prototype database 115 is modeled using ellipses, further improvements in recognition performance can be achieved. Therefore, instead of the radius as described above, the principal axis diameter of the scattering ellipse and its tilt relative to the coordinate system must now be stored in the prototype database 115 for each preferably signal basic object prototype in the prototype database 115. The disadvantage is a significant further increase in computational power and storage requirements.

[0252] Of course, the computation can be made more complex, but this usually only increases the cost significantly and the performance of the signal basic object prototype recognizer in the prototype database 115 is no longer significantly improved.

[0253] Therefore, it is recommended to use the simplest of the described approaches.

[0254] Now, by determining the distance, 112 is... Figure 8The orientation of the current eigenvector signal value of eigenvector signal 138, determined in the exemplary two-dimensional parameter space, may be very different. It is conceivable that this first eigenvector signal value 146 is too far from the centroid coordinates 141, 142, 143, 144 of the centroid of any signal primitive object prototype in the prototype database 115. This distance threshold could, for example, be half of the mentioned minimum prototype distance. It is also possible that the signal primitive object prototypes overlap in the scattering range around their respective centroids 143, 142, and the second current eigenvector signal value 145 of eigenvector signal 138, for example, is within this overlap. In this case, it is assumed that the list may contain two signal primitive object prototypes with different probabilities as additional parameters (for different distances). That is, instead of handing over the most probable signal primitive object to the Viterbi estimator 113, the vector consisting of the possible signal primitive objects is handed over to the Viterbi estimator 113. Next, from the time series of these hypothesis lists, the Viterbi estimator searches for possible sequences having the highest probability of one of the pre-given signal basic object sequences in its signal object database relative to all possible paths traversing the hypothesis lists received from the distance determiner 112 via the Viterbi estimator 113 for identified signal basic object 121. Here, for each hypothesis list, exactly one identified signal basic object prototype from that hypothesis list must traverse that path.

[0255] In the best case, the current eigenvector signal value 148 is within the scattering range (threshold ellipsoid) 147 around the centroid 141 of the unique signal basic object prototype 141, which is thus reliably identified by the distance determiner 112 and handed over to the Viterbi estimator 113 as the identified signal basic object 121.

[0256] It is conceivable that, in order to improve the modeling of the dispersion range of a single signal basic object prototype, this dispersion range could be modeled using multiple signal basic object prototypes, which are circular in shape and each possesses its own dispersion range. That is, multiple signal basic object prototypes in prototype database 115 can represent the same signal basic object prototype in the understanding of signal basic object categories. The danger here is that, because the probability of a signal basic object prototype is distributed across multiple such sub-signal basic object prototypes, the probability of a single sub-signal basic object prototype might become smaller than the probability of other signal basic object prototypes whose probability is less than that of the original signal basic object prototype. Therefore, these other signal basic object prototypes might be incorrectly constructed.

[0257] Another important issue is computing power, which must be provided to reliably identify the signal basic object prototypes in prototype database 115. This point deserves further discussion:

[0258] The key point is: computational costs increase with... That is, it increases with the number and dimensions of the codebook.

[0259] In the case of an unoptimized HMM recognizer, the number of assembly instructions that must be implemented to compute the vector components is approximately eight steps.

[0260] The number of assembly steps, A_Abst, required to calculate the distance from a single signal primitive object prototype CbE in the prototype database to a single eigenvector signal value FV is approximately calculated as follows:

[0261]

[0262] This results in an A_CB number of assembly steps used to determine the signal basic object prototype with the minimum distance in prototype database 115:

[0263]

[0264] Taking a medium-sized HMM recognizer with 50,000 basic signal object prototypes (the number of basic signal object prototype entries in the prototype database = CB_anz) and 24 FV_Dimensions (the number of parameter values ​​in the eigenvector signal values ​​= eigenvector dimension = FV_Dimension) as an example, the number of steps is:

[0265] Each eigenvector signal value of eigenvector signal 138 has Millions of operations

[0266] At a relatively low sampling rate of 8kHz = 8000 FV per second (124 eigenvector signal values ​​per second), a computing power of 8GIpS (8 billion instructions per second) is already required.

[0267] Given the challenges of energy conservation and / or CO2 reduction for electric vehicles, this is unacceptable.

[0268] Therefore, as already mentioned, when performing the optimized HMM recognition method via distance determiner 112 or classifier 112, the minimum distance between two basic signal object prototypes in prototype database 115 is pre-calculated and stored in prototype database 115 or in distance calculation 112. This has the advantage that if distance determiner 115 finds a distance less than half of the minimum distance between the current feature vector signal value of feature vector signal 138 and the basic signal object prototype in prototype database 115, distance determiner 112 can interrupt the search. This halves the average search time for distance determiner 112. Further optimizations can be made if prototype database 115 is sorted according to the statistical occurrence of basic signal object prototypes in real ultrasonic echo signal 1. This ensures that the most frequent basic signal object prototypes can be found much faster, further reducing the computation time of distance determiner 112 or classifier 112 and further reducing power consumption.

[0269] For the distance determiner implementing the therefore optimized HMM recognition process, the computational requirements are now as follows:

[0270] Next, there are 8 steps used to calculate the distance between vector components. The steps used to calculate the distance A_Abst between the signal primitive object prototype entry CbE in prototype database 115 and the current eigenvector signal value FV of eigenvector signal 138 are:

[0271]

[0272] The number of steps used to determine the signal basic object prototype entries with the minimum distance A_CB in the prototype database 115 under optimized conditions is slightly higher:

[0273]

[0274] Two additional assembly instructions are required to check whether the distance between the determined current eigenvector signal value and the just-checked signal basic object prototype in prototype database 115 is less than half of the minimum distance between signal basic object prototypes in prototype database 115.

[0275] Furthermore, the number of signal basic object prototype entries CB_Anz in the prototype database 115 for mobile and energy self-sufficient applications is limited to 4,000 prototype database entries of signal basic object prototypes in the prototype database 115 or even less.

[0276] Furthermore, the number of feature vector signal values ​​per second within the feature vector signal 138 is reduced by filtering in the feature vector extractor 111 and by reducing the sampling rate in the feature vector extractor 111.

[0277] This is illustrated with a simple example:

[0278] The distance determiner 112 or classifier 112 mentioned in the implementation of the intermediate HMM recognition method is now operated using a prototype database 115 that still has less than one-tenth of the entries, for example, 4000 entries CbE and 24 feature vector signal dimensions (i.e. 24 parameter signals).

[0279] Now, the number of steps is

[0280] Each eigenvector signal value of eigenvector signal 138 has Second operation

[0281] When the eigenvector signal value rate decreases to 100 eigenvector signal values ​​per second, more than, for example, 80 sample values ​​are extracted from a 10ms time window in the eigenvector extractor 111, and the search is interrupted when the distance between the current eigenvector signal value and the processed signal basic object prototype of the prototype database 115 is less than half the distance of the minimum prototype database entry, the cost is reduced by at least half with proper sorting of the prototype database 115.

[0282] Next, the required computing power is reduced to <33 DSP-Mips (33 million operations per second). In fact, the prototype database sorting results in an even lower computing power requirement, for example, 30 Mips. This allows the system to achieve real-time capabilities and be integrated into a single IC, thereby enabling full integration into the sensor.

[0283] Pre-selection can limit the search space. Here, the premise is that the data is uniformly distributed, and the centroid of the quadrant is at the center of the geometric quadrant.

[0284] The necessary reduction in the size of the prototype database 115 has the following advantages and disadvantages:

[0285] Reducing the number of entries in the prototype database 115 not only improves the false acceptance rate (FAR), i.e., the false signal basic object prototypes that are accepted as signal basic object prototypes, but also improves the false rejection rate (FRR), i.e., the actual signal basic object prototypes that are not identified.

[0286] On the other hand, this reduces resource requirements (computing power, chip area, memory, power consumption, etc.).

[0287] Furthermore, when forming hypotheses through the distance determiner 112, previous history, that is, previously identified signal prototypes, can be used. A suitable model for this is the so-called Hidden Markov Model (HMM).

[0288] Therefore, for each basic signal prototype, a confidence level and a distance from the measured current eigenvector signal value of the eigenvector signal 138 can be derived, which can also be further processed by the Viterbi estimator 113. It is also reasonable to output a hypothesis list for each identified basic signal prototype 121, which, for example, contains ten most probable basic signal prototypes with the corresponding probability and reliability of the identification.

[0289] Since not every temporal and spatial signal basic object sequence can be assigned to a signal object, it is possible to evaluate a list of hypothesis sequences of consecutive frames using the Viterbi estimator 113.

[0290] In this case, the basic object prototype sequence path of the signal can be found by a continuous hypothesis list, which has the highest probability and exists in the signal object database 116 of the Viterbi estimator 113.

[0291] In this case, at least two identifications are also performed:

[0292] 1. Is the most likely sequence of the basic object prototype of a signal one of the already stored sequences of the basic object prototype of a signal, and with what probability and reliability is it one of the already stored sequences of the basic object prototype of a signal?

[0293] 2. If the sequence is one of the already stored sequences of the basic object prototype of the signal, then which sequence is it and with what probability and reliability is it?

[0294] Therefore, on the one hand, such prototype signals in the form of entries consisting of a pre-given sequence of basic signal object prototypes can be input into the object database 116 through a learning program; on the other hand, this can also be done manually via an input tool that enables the input of these sequences of basic signal object prototypes via a keyboard.

[0295] With the aid of the Viterbi estimator 113, the most probable sequence of signal basic object prototypes in a predefined sequence can be determined based on the sequence of the hypothesis list of the distance determiner 112 or classifier 112 for the identified sequence of signal basic object prototypes 121. This is especially applicable when individual signal basic object prototypes are misidentified due to measurement errors caused by the distance determiner 112 or classifier 112. In this respect, it is quite reasonable for the Viterbi estimator 113 to take over the sequence of the hypothesis list of signal basic object prototypes of the distance determiner 112 or classifier 112, as described above. The result is the signal object 122 that is identified as the most probable, or the signal object hypothesis list, similar to the emission calculation of the distance estimator 112 or classifier 112 described previously.

[0296] Finally, the functional components of the signal object recognition machine should be considered. These functional components... Figure 7 The Viterbi estimator 113 is incorporated into this process. This Viterbi estimator 113 searches and accesses a signal object database 116. The signal object database 116 is fed both through a learning tool and through a tool, wherein these sequences of basic signal object prototypes can be specified via text input. The possibility of downloading in production should be mentioned here only for completeness.

[0297] In the Viterbi estimator 119, the illustrative basis for sequence identification of time series of signal basic object prototypes is, exemplarily, a Hidden Markov Model. This model is built from different states. Figure 9 In the example illustrated, these states are symbolically represented by numbered circles. Figure 9 In the example mentioned, these circles are numbered from Z1 to Z6. Transitions are formed between these states. These transitions are... Figure 9 The sequence is represented by the letter 'a' and two indices 'i' and 'j'. The first index 'i' represents the output node number, and the second index 'j' represents the target node number. Besides transitions between two different nodes, there are also transitions that return to the starting node, either 'aii' or 'ajj'. There are also transitions that can skip nodes. Therefore, based on this sequence, the probability of actually observing the k-th observable 'bk' is obtained. Thus, the observable sequence that can be observed using a pre-calculated probability 'bk' is obtained.

[0298] Importantly, every Hidden Markov Model consists of unobservable states q. i Composition. In two states q i With q j Between them, there exists a transition probability a ij .

[0299] Therefore, used from q iTransfer to q j The probability p can be written as:

[0300]

[0301] Here, n represents a discrete time point. That is, in a state q i Step n and having state q j The transition occurs between steps n+1. Emission distribution b i (Ge) depends on state q i As already explained, this is in the system (Hidden Markov Model) at state q. i The probability of observing (observable) the most basic Ge:

[0302]

[0303] In order to start the system, an initial state must be defined. This is done through the probability vector π. i This can be achieved. Then it can be explained that the probability is π. i The initial state is qi:

[0304]

[0305] Importantly, a new model must be created for each sequence of the signal's basic object prototype. In model M, the observation probabilities of the time-observed sequences of the signal's basic object prototype should be determined.

[0306]

[0307] This corresponds to a time-state sequence that is not directly observable, and that corresponds to the following sequence:

[0308]

[0309] The probability p of observing the time state sequence Q, which depends on the model M, the state sequence Q, and the time observation sequence Ge, is:

[0310]

[0311] This serves as a state sequence. The probability of this is obtained in model M:

[0312]

[0313] Therefore, the probability used to identify the signal object is the same as the sequence of the basic object prototype of the signal (see also...). Figure 10 ):

[0314]

[0315] Here, a single probability Q is used to determine all possible paths of the observed sequence that cause the basic object prototype Ge of the signal. k This summation is used to determine the most likely signal object model (signal object) of the observed emitted Ge.

[0316]

[0317] Summing all possible paths Q is problematic due to potential computational costs, and is therefore often interrupted very early. Therefore, it is proposed to use only the most probable path Q. k This point will be discussed below.

[0318] This calculation is performed recursively. At time point n, the system is observed to be in state q. i The probability an(i) can be calculated as follows:

[0319]

[0320] In this case, we enter state q. i+1 Summate all S possible paths.

[0321] Here we assume: reaching state q i n+1 The probability is determined by the optimal path. Therefore, the sum can be simplified with low error.

[0322]

[0323] Now, by backtracking from the last state, we can obtain the optimal path.

[0324] The probability of this path is a mathematical product. Therefore, logarithmic calculation simplifies the problem to a pure summation problem. Here, it is used to identify the signal object (which corresponds to model M). j The probability of identification corresponds to the determination of the most likely signal object model for the observed emission X. Now, the emission only travels through the optimal possible path Q. best To achieve

[0325]

[0326] Therefore, the probability becomes

[0327]

[0328] What is particularly important now is that prototype database 115 only contains prototypes of basic signal objects.

[0329] The identified signal object is transmitted along with the identified parameters, rather than along with the sampled values. This results in data compression without discarding signal characteristics.

[0330] Of particular importance: obstacle objects within the vehicle's environment are not identified here. More precisely, structures within the ultrasonic echo signal 1 are identified and used for compression.

[0331] Only in this way can the signal be reconstructed without evaluation in the controller after receiving data.

[0332] That is, unlike the prior art, the objective of the present invention is not to detect obstacles in the vehicle environment and classify these obstacles and thereby cause data compression, but to compress and transmit the ultrasonic echo signal 1 itself as losslessly as possible by limiting it to signal shape components relevant to the application.

[0333] Next, the ultrasonic sensor transmits compressed data to the computer system, preferably only transmitting the codes (symbols) of the identified prototypes, the amplitude and / or time extension of these prototypes, and the timestamps of their occurrence. This minimizes the EMV load caused by data transmission via the data bus between the ultrasonic sensor and the computer system, and allows the ultrasonic sensor's status data to be transmitted to the computer system via the data bus for system error identification within time intervals, improving latency. In conjunction with this invention, it has been recognized that prioritizing data transmissions via the data bus can be advantageous. However, this prioritization does not involve prioritization relative to other bus users as is known from the prior art. More precisely, the data connection between the vehicle's sensors and the computer system is typically a point-to-point connection. Therefore, more precisely, this prioritization can be understood as: which data determined by the sensor system must be transmitted as the first controller in time. Here, reporting sensor safety-critical errors to the computer system has the highest priority, as these errors have a high probability of compromising the validity of the ultrasonic sensor's measurement data. This data is sent from the sensors to the computer system. Requests to perform safety-related self-tests have the second highest priority from the computer system. These instructions are then sent from the computer system to the sensors. Since latency cannot be increased, the data from the ultrasonic sensors themselves has the third highest priority. All other data has a lower priority for transmissions via the data bus.

[0334] Particularly advantageous is the design of a method for transmitting sensor data, particularly ultrasonic sensor data, from a sensor to a computer system, especially a computer system in a vehicle, the method comprising emitting an ultrasonic pulse train at a start 57 and an end 56 and including a reception time T from at least the end 56 of the emission of the ultrasonic pulse train. E The system internally generates a received signal and transmits compressed data to the computer system via a data bus, particularly a single-wire data bus. This transmission of data from the sensor 54 to the computer system begins either after the start command 53 from the computer system to the ultrasonic sensor via the data bus and before the end 56 of the ultrasonic pulse train, or after the start command 53 from the computer system to the sensor via the data bus and before the start 57 of the ultrasonic pulse train. Then, transmission 54 continues periodically after the start command 53 until the data transmission 58 ends. The end of the data transmission 58 occurs at the receiving time T. E After it's over.

[0335] Therefore, as the first step in data compression, another variation of the proposed method specifies that a feature vector signal (a stream of feature vectors with n feature vector values ​​and n as the dimension of the feature vectors) is formed based on the received signal. This feature vector signal can include multiple analog and digital data signals. That is, the feature vector signal is a time series of data / signal structures that are more or less complex. In its simplest case, the feature vector signal can be understood as a vector signal composed of multiple partial signals.

[0336] For example, it might be reasonable to calculate the first and / or higher-order time derivatives of the received signal, or the single or multiple integrals of the received signal, then these time derivatives and integrals are portions of the signal within the eigenvector signal.

[0337] It can also form an envelope signal of the received signal (ultrasonic echo signal), and this envelope signal can be a part of the signal within the characteristic vector signal 138.

[0338] Another plausible approach is to convolve the received signal with the emitted ultrasonic signal to form a correlation signal, which could be a portion of the eigenvector signal. This correlation signal could then be used, on the one hand, as the emitted ultrasonic signal used to control the transmitter's driver, or on the other hand, a signal measured at the transmitter that better corresponds to the actual emitted sound wave could be used.

[0339] Finally, it might be reasonable to use an optimal filter to detect the presence of predetermined signal objects and to form optimally filtered signals for corresponding signal objects among some of these predetermined signal objects. In this case, the optimal filter is a filter optimized for the signal-to-noise ratio (SNR). In the interfered ultrasound received signal, predefined signal objects should be identified. Name-related filters, signal-matched filters (SAFs), or simply matched filters also frequently appear in this literature. The optimal filter is used to optimally determine (detect) (parameter estimation) the presence of known signal shapes, i.e., the amplitude and / or azimuth of predetermined signal objects, in the presence of interference. These interferences might be, for example, signals from other ultrasound transmitters and / or ground echoes.

[0340] Therefore, the optimal filter output signal is preferably a portion of the signal within the eigenvector signal.

[0341] Certain events can be signaled in a separate portion of the eigenvector signal. These events are signal fundamental objects for the purposes of this invention. That is, signal fundamental objects do not include signal forms such as rectangular pulses or wavelets or wave trains, but rather include unique points in the waveform of the received signal and / or in the waveform of the signal derived therefrom, such as envelope signals (ultrasonic echo signals), which can be obtained, for example, by filtering from the received signal.

[0342] Another signal, which can be a part of the eigenvector signal, can be used to indicate, for example, whether the envelope of the received signal or the ultrasonic echo signal crosses a predetermined third threshold. That is, it involves a signal that indicates the presence of a signal fundamental object within the received signal and thus the eigenvector signal.

[0343] Another signal, which can be a part of the eigenvector signal, can be used to indicate, for example, whether the envelope of the received signal or the ultrasonic echo signal rises above a predetermined fourth threshold, which may be the same as the third threshold. That is, it involves a signal that indicates the presence of a signal fundamental object within the received signal and thereby the eigenvector signal.

[0344] Another signal, which can be a part of the eigenvector signal, can be used to indicate, for example, whether the envelope of the received signal, or the ultrasonic echo signal, droops beyond a predetermined fifth threshold, which may be the same as a third or fourth threshold. That is, it involves a signal that indicates the presence of a signal fundamental object within the received signal and thus the eigenvector signal.

[0345] Another signal, which can be a part of the eigenvector signal, can be used to indicate, for example, whether the envelope of the received signal or the ultrasonic echo signal has a maximum value higher than a sixth threshold, which can be the same as the third to fifth thresholds mentioned earlier. That is, it involves a signal that indicates the presence of a signal fundamental object within the received signal and thereby the eigenvector signal.

[0346] Another signal, which can be a part of the eigenvector signal, can be used to indicate, for example, whether the envelope of the received signal or the ultrasonic echo signal has a minimum value higher than a seventh threshold, which can be the same as the third to sixth thresholds mentioned earlier. That is, it involves a signal that indicates the presence of a signal fundamental object within the received signal and thereby the eigenvector signal.

[0347] Here, it is preferably evaluated that at least one previous maximum value of the ultrasonic echo signal has a minimum distance from the minimum value in order to avoid detecting noise. Other filtering is conceivable in this regard. It can also be checked that the time interval between the minimum value and the previous maximum value is greater than a first minimum time interval. The satisfaction of these conditions sets a flag or signal, which itself is preferably a portion of the eigenvector signal.

[0348] Similarly, the time intervals and amplitude distances of other signal objects should be checked to ensure they meet certain reasonable requirements, such as minimum time intervals and / or minimum amplitude distances. Based on these checks, other partial signals, including analog, binary, or digital signals, can also be derived, thus further increasing the dimension of the eigenvector signal.

[0349] If necessary, the eigenvector signal can be transformed into a significant eigenvector signal during the significance enhancement stage, for example, through linear mapping or higher-order matrix polynomials. However, practice has shown that this is not necessary, at least for current requirements.

[0350] According to the proposed method, signal objects are then identified within the received signal based on eigenvector signals or significant eigenvector signals, and these signal objects are classified into the identified signal object categories.

[0351] If, for example, the output signal of an optimal filter, and by virtue of a portion of the eigenvector signal, has an amplitude higher than, if necessary, an eighth threshold specific to the optimal filter, then the signal object for which the optimal filter is designed to detect can be considered identified. Here, other parameters are also preferably considered. If, for example, an ultrasonic pulse train is transmitted at a frequency that rises during the pulse train (called a chirp), then an echo with this modulation characteristic is also expected. If the signal shape of the ultrasonic echo signal, for example, the triangular signal shape of the ultrasonic echo signal, is temporally consistent with the expected signal shape, but is not a modulation characteristic, then it is not an echo from the transmitter, but an interference signal, which may originate from another ultrasonic transmitter or from the ultrasonic propagation distance. In this respect, the system can then distinguish between self-echoes and incoming echoes, thereby assigning the same signal shape to two different signal objects, namely self-echoes and incoming echoes. Here, it is preferable to prioritize the transmission of self-echoes from the sensor to the computer system via the data bus compared to the transmission of incoming echoes, because the former is generally security-related while the latter is generally security-unrelated.

[0352] Typically, at least one signal object parameter is assigned to each identified signal object, or at least one signal object parameter is determined for that signal object. Preferably, the signal object parameter is a timestamp indicating when the signal object was received. Here, the timestamp may, for example, relate to the start time of the signal object in the received signal or the time orientation of the signal object's time centroid, etc. Other signal object parameters, such as amplitude, extension, etc., are also conceivable. Thus, in a variation of the method according to the invention, at least one of the assigned signal object parameters is transmitted along with a symbol for at least one identified signal object category. Preferably, the signal object parameter is a time value as a timestamp and indicates a time position suitable for inferring the (reception) time since the previous ultrasound pulse train was emitted. Preferably, the determined distance of the object is later determined based on this determined and transmitted time value.

[0353] Finally, preferably, the identified signal object categories are transmitted in a priority order, using assigned timestamped symbols, together with the assigned signal object parameters. This transmission can also be implemented in more complex data structures (Records). For example, it is conceivable to first transmit the time point of the identified safety-related signal objects (e.g., identified obstacles) and then transmit the identified signal object categories of the safety-related signal objects. This further reduces latency.

[0354] In one variant, the method according to the invention includes at least: determining a chirp value as an assigned signal object parameter, the chirp value indicating whether the identified signal object is an echo of an ultrasonic transmission pulse train with chirp-up, chirp-down, or chirp-free characteristics. Chirp-up means that the frequency of the received signal within the signal object increases. Chirp-down means that the frequency of the received signal within the signal object decreases. Chirp-free means that the frequency of the received signal within the signal object remains substantially constant. In this regard, reference should be made to DE-B-10 2017 123 049, DE-B-10 2017 123 050, DE-B-10 2017 123 051, and DE-B-10 2017 123 052, the entire contents of which are part of this invention.

[0355] Therefore, in a variation of the method according to the invention, a confidence signal is also formed by receiving a signal on one side, or by replacing the received signal with a signal derived from the received signal, and on the other side a reference signal, such as an ultrasonic transmission signal or another anticipated wavelet, forming a correlation, for example, a time-continuous or time-discrete correlation integral. The confidence signal is typically a partial signal of the eigenvector signal, i.e., a component of the eigenvector consisting of a sequence of vector sample values ​​(eigenvector values).

[0356] In a variation of the method according to the invention, a phase signal is also formed based on this, which describes, for example, the phase shift of a received signal or a signal formed therefrom (e.g., a confidence signal) relative to a reference signal, such as an ultrasonic transmission signal and / or other reference signals. Thus, the phase signal is typically also a partial signal of the eigenvector signal, i.e., a component of the eigenvector consisting of a sequence of vector sample values.

[0357] Similarly, in another variation of the method according to the invention, a phase position signal can be formed by establishing a correlation between a phase signal or a signal derived therefrom and a reference signal, and this phase position signal can be used as a part of the eigenvector signal. Thus, the phase position signal is typically also a part of the eigenvector signal, i.e., a component of the eigenvector consisting of a sequence of vector sample values.

[0358] Therefore, when evaluating the eigenvector signal, it might be reasonable to perform a comparison of the phase position signal with one or more thresholds to generate a discretized phase position signal, which itself can become a part of the eigenvector signal.

[0359] In a variation of the proposed method, eigenvector signals and / or salient eigenvector signals can be evaluated such that one or more distance values ​​are formed between the eigenvector signals and one or more signal object prototype values ​​for identifiable signal object categories. These distance values ​​can be Boolean, binary, discrete, digital, or analog. Preferably, all distance values ​​are correlated with each other in a nonlinear function. Thus, in the case of a anticipated triangular chirped upward echo, the received triangular chirped downward echo can be discarded. This discarding is a nonlinear process for the purposes of this invention.

[0360] Conversely, the triangle in the received signal can behave differently. This primarily involves the amplitude of the triangle in the received signal. If the amplitude in the received signal is sufficient, then, for example, the optimal filter assigned to that triangle signal provides a signal above a pre-given ninth threshold. Then, in this case, for example, the signal object category (triangle signal) can be assigned to the signal object identified at the specified time point. In this case, the distance value between the feature vector signal and the prototype (here, the ninth threshold) is lower than one or more pre-determined, binary, digital, or analog distance values ​​(here, 0 = crossover).

[0361] In another variation of the method according to the invention, at least one category of signal objects is wavelets, which are estimated and thereby detected by estimation devices (e.g., optimal filters) and / or estimation methods (e.g., estimation programs running in a digital signal processor). The term "wavelet" refers to a function that can be based on continuous or discrete wavelet transforms. The word "wavelet" is a neologism from the French "ondelette," which means "small wave" and is adopted into English partly literally ("onde" -> "wave") and partly phonetically ("-lette" -> "-let"). The term "wavelet" was coined in the 1980s in geophysics (Jean Morlet, Alex Grossmann) to summarize functions of the short-term Fourier transform, but its current conventional meaning has only been used since the late 1980s. In the 1990s, the discovery of compact, continuous (up to any differentiable order) and orthogonal wavelets by Ingrid Daubechies (1988) and the development of the Fast Wavelet Transform (FWT) algorithm by Stéphane Mallat and Yves Meyer (1989) using MultiResolution Analysis (MRA) sparked a real wavelet boom.

[0362] Unlike the sine and cosine functions of the Fourier transform, the most commonly used wavelets exhibit locality not only in the frequency domain but also in the time domain. Here, "locality" should be understood as propagation within a small area. The probability density is the normalized numerical square of the function under consideration or its Fourier transform. Here, the product of the two variances is always greater than a constant, similar to Heisenberg's uncertainty principle. Based on this constraint, the Paley-Wiener theory (Raymond Paley, Norbert Wiener) – a precursor to the discrete wavelet transform – was developed in function analysis. theory( (Antoni Zygmund), this theory corresponds to the continuous wavelet transform.

[0363] While the integral of a wavelet function is always zero, wavelet functions typically take the form of outward-running (shrinking) waves (i.e., "Wellchen" = Ondelettes = wavelet). However, for the purposes of this invention, wavelets with non-zero integrals should also be permitted. Here, rectangular and triangular wavelets described below should be mentioned exemplarily. This other interpretation of the term "wavelet" is popular in American English-speaking regions and is therefore well-known. This other interpretation should also apply.

[0364] Important examples of wavelets with an integral of 0 are the Haar wavelet (Alfréd Haar, 1909), the Daubechies wavelet named after Ingrid Daubechies (circa 1990), the Coiflet wavelet also constructed by her, and the theoretically more important Meyer wavelet (Yves Meyer, circa 1988).

[0365] Wavelets exist for any dimensional space, and most are tensor products of one-dimensional wavelet bases. Due to the fractal nature of the two-scale equations in MRA, most wavelets have complex shapes, which are mostly not closed. This is particularly important because the eigenvector signals mentioned earlier are multidimensional, thus allowing the use of multidimensional wavelets for signal object recognition.

[0366] Therefore, a particular variation of the proposed method is to use multidimensional wavelets with more than two dimensions for signal object recognition. In particular, it is proposed to use corresponding optimal filters to recognize such wavelets with more than two dimensions, so as to supplement the feature vector signal with other parts of the signal suitable for recognition when necessary.

[0367] A particularly suitable wavelet is the triangular wavelet. The characteristics of the triangular wavelet are: a starting time point, a wavelet amplitude that rises substantially linearly in time after the starting time point until the maximum amplitude, and a wavelet amplitude that falls substantially linearly in time after the maximum amplitude of the triangular wavelet until the end of the triangular wavelet.

[0368] Another particularly suitable wavelet is the rectangular wavelet, which, for the purposes of this invention, also includes the trapezoidal wavelet. The rectangular wavelet is characterized by a start time point after which the wavelet amplitude increases with a first time steepness until a first plateau time point. After the first plateau time point, the wavelet amplitude maintains a second time steepness until a second plateau time point. After the second plateau time point, it decreases with a third time steepness until the end of the rectangular wavelet's time. Here, the value of the second time steepness is less than 10% of the value of the first time steepness and less than 10% of the value of the third time steepness.

[0369] Instead of the wavelets described above, other two-dimensional wavelets, such as the sinusoidal half-wave wavelet, can also be used, which also has an integral that is not equal to zero.

[0370] The proposal suggests that when using wavelets, the time offset of the correlated wavelet of the identified signal object is used as a signal object parameter. For example, this offset can be determined through correlation. Further, it is proposed that when using wavelets, the output level of an optimal filter suitable for detecting the correlated wavelet is preferably used at a time point where the level exceeds a predefined tenth threshold for the signal object or the wavelet. Preferably, the ultrasonic echo signal (envelope) and / or phase signal and / or confidence signal, etc., of the received signal are evaluated.

[0371] Another possible parameter of the signal object that can be determined is the time compression or spread of the associated wavelet of the identified signal object. Similarly, the amplitude of the wavelet of the identified signal object can also be determined.

[0372] Within the framework of this invention, it has been recognized that it is advantageous to transmit data of the identified signal objects whose echoes arrive very quickly from the sensor to the computer system first, and then transmit subsequent data of the signal objects identified later from the sensor to the computer system. Preferably, at least the category of the identified signal object and a timestamp, which preferably indicates the time when the signal object re-arrives at the sensor, are always transmitted. Within the framework of the identification process, different signal objects considered for receiving the signal can be assigned score values, which indicate the probability assigned to the presence of the signal object according to the estimation algorithm used. In its simplest case, such score values ​​are binary values. However, preferably, the scores are complex numbers, real numbers, or integers. For example, they may relate to a determined distance. In some cases, it is reasonable to also transmit data of identified signal objects with lower score values, provided that multiple signal objects have high score values. For the computer system to operate correctly, in such cases, not only the data (symbols) of the identified signal objects and the corresponding timestamps of the signal objects should be transmitted, but also the determined score values. Instead of transmitting only the data (symbols) of the identified signal objects and the timestamps of the signal objects corresponding to those symbols, it is also possible to additionally transmit the data (symbols) of signal objects with a second smallest distance and the timestamps of the signal objects corresponding to those second possible symbols. Therefore, in this case, a hypothesis list consisting of two identified signal objects, their time positions, and additionally assigned score values ​​is transmitted to the computer system. Of course, a hypothesis list consisting of more than two symbols for more than two identified signal objects, their time positions, and additionally assigned score values ​​is also transmitted to the computer system.

[0373] Preferably, the data for the identified signal object category and the assigned data, such as the timestamp and score value of the corresponding identified signal object category, are transmitted according to the First-In-First-Out (FIFO) principle. This ensures that the data of the reflection of the nearest object is always transmitted first, and that safety-critical situations involving vehicle-obstacle collisions are handled in a probabilistic priority order.

[0374] In addition to transmitting measurement data, fault data of the sensor can also be transmitted. If the sensor is found to be damaged by a self-test device and previously transmitted data may potentially be erroneous, this can also be transmitted at the receiving time T. EThis is done during the process. Therefore, it ensures that the computer system can be aware of changes in the evaluation of the measurement data at the earliest possible time and can discard or process this measurement data accordingly. This is important for emergency braking systems, as emergency braking is a safety-critical intervention that is only permitted to be introduced if the data on which it is based has a corresponding level of confidence. Therefore, correspondingly, the transmission of measurement data, i.e., data on the identified signal object category, and / or parameters of at least one assigned signal object parameter, is postponed and thus given a lower priority. Of course, it is conceivable to interrupt transmission in the event of a sensor malfunction. However, in some cases, it is also possible that a malfunction may occur but not necessarily exist. In this case, it may be appropriate to continue transmission. Therefore, it is preferable to transmit sensor safety-critical malfunctions with a higher priority.

[0375] In addition to the wavelets with an integral value of 0 already described and the signal portions of wavelets with an integral value of non-zero, which are additionally referred to herein as wavelets, certain time points during the signal reception process can also be understood as signal basic objects for the purposes of this invention. These time points can be used for data compression and can be transmitted in place of the sampled values ​​of the received signal. This subset of the possible set of signal basic objects is hereinafter referred to as signal time points. That is, for the purposes of this invention, these signal time points are a special form of signal basic objects.

[0376] The first possible signal time point, and by virtue of this signal, is the intersection of the amplitude of the ultrasonic echo signal 1 and the amplitude of the eleventh threshold signal SW in the rising direction.

[0377] The second possible signal time point, and by virtue of this signal, is the intersection of the amplitude of the ultrasonic echo signal 1 and the amplitude of the twelfth threshold signal SW in the descending direction.

[0378] The third possible signal time point, and by virtue of this signal basic object, is the maximum value of the amplitude of the ultrasonic echo signal 1 that is higher than the amplitude of the thirteenth threshold signal SW.

[0379] The fourth possible signal time point, and by virtue of this signal basic object, is the minimum value of the amplitude of the ultrasonic echo signal 1 that is higher than the amplitude of the fourteenth threshold signal SW.

[0380] If necessary, it may be reasonable to use a threshold signal SW specific to the signal time point type for these four exemplary signal time point types and for other signal time point types.

[0381] The time series of the basic signal objects are generally not arbitrary. This is advantageously used because, preferably, simpler basic signal objects should not be transmitted, but rather the identified patterns of the sequences of these basic signal objects that represent the actual signal objects. If, for example, a triangular wavelet is expected in an ultrasonic echo signal 1 of sufficient amplitude, then, in addition to the corresponding minimum level at the output of the optimal filter suitable for detecting such a triangular wavelet, a time period exceeding the mentioned minimum level at the output of the aforementioned optimal filter can be expected.

[0382] 1. The first possible signal time point where the amplitude of the ultrasonic echo signal 1 crosses the amplitude of the threshold signal SW in the rising direction, and the time point immediately following this point.

[0383] 2. A third possible signal time point that occurs with an amplitude of ultrasonic echo signal 1 that is higher than the maximum amplitude of the threshold signal SW, and which is immediately following this point in time.

[0384] 3. A second possible signal time point in which the amplitude of the ultrasonic echo signal 1 and the amplitude of the threshold signal SW intersect in the descending direction.

[0385] Thus, in this example, the signal object of the triangular wavelet lies in a predefined sequence of three basic signal objects, which are identified and replaced with symbols and are preferably transmitted along with their occurrence time, i.e., timestamps. Furthermore, this exceeding the minimum level mentioned at the output of the aforementioned optimal filter is the fifth possible signal time point and thus another possible basic signal object.

[0386] Therefore, the resulting grouping and time series of the identified signal fundamental objects can themselves be identified, for example, by a Viterbi decoder as predefined, expected groupings or time series of signal fundamental objects, and thus can themselves represent signal fundamental objects. Therefore, the sixth possible signal time point, and thus the signal fundamental object, is the occurrence of such predefined groupings and / or time series of other signal fundamental objects.

[0387] If such grouping of signal basic objects or time series of such signal object categories in the form of signal objects are identified, it is preferable to transmit the symbols of the identified aggregate signal object category and at least the assigned signal object parameters instead of transmitting the individual signal basic objects. However, it is also possible that both are transmitted. In this case, data (symbols) of the signal object category of the signal objects, which are predefined time series and / or groupings of other signal basic objects, are transmitted. For compression purposes, it is advantageous not to transmit at least one signal object category (symbol) of at least one of these other signal basic objects.

[0388] In particular, time grouping of signal basic objects exists when the time intervals between these basic objects do not exceed a predefined interval. In the previously mentioned example, the transit time of the signal in the optimal filter should be considered. Typically, the optimal filter allows for a slower transit time than the comparator. Therefore, the variation in the output signal of the optimal filter should maintain a fixed temporal relationship with the time occurrence of important signal points.

[0389] This paper proposes a method for transmitting sensor data, particularly ultrasonic sensor data, from a sensor to a computer system, especially a computer system in a vehicle. The method begins with emitting an ultrasonic pulse train and receiving an ultrasonic signal, forming a time-discrete received signal consisting of a sequence of sampled values. Each sampled value is assigned a date (timestamp). Then, using at least one suitable filter (e.g., an optimal filter), at least two parameter signals are determined from the sequence of sampled values ​​of the received signal, each relating to the existence of a signal fundamental object assigned to its respective parameter signal. The resulting parameter signals (eigenvector signals) are also designed as time-discrete sequences of corresponding parameter signal values ​​(eigenvector values), each associated with a date (timestamp). Therefore, it is preferable to assign exactly one date (timestamp) to each parameter signal value (eigenvector value). Hereinafter, these parameter signals are collectively referred to as eigenvector signals. Thus, the eigenvector signals are designed as time-discrete sequences of eigenvector signal values, each having n parameter signal values, which are composed of these parameter signal values ​​and other parameter signal values ​​each having the same date (timestamp). In this case, n is the dimension of the individual eigenvector signal values, which preferably remain the same from one eigenvector value to the next. Each such eigenvector signal value is assigned a corresponding time date (timestamp). The temporal waveform of the eigenvector signal in the resulting n-dimensional phase space is then evaluated, and the identified signal object is inferred based on a determined evaluation value (e.g., distance). As mentioned above, the signal object here consists of a time sequence of basic signal objects. Here, the signal object is typically assigned a symbol beforehand. In other words, it is checked whether the point pointed to by the n-dimensional eigenvector signal in the n-dimensional phase space is closer than a predetermined point in the n-dimensional phase space by a predetermined maximum distance along its path through the n-dimensional phase space in a predetermined time order. That is, the eigenvector signal has a temporal waveform. Next, an evaluation value (e.g., distance) is calculated, which may reflect, for example, the probability of the existence of a particular sequence. Then, the evaluation value, reassigned with a time date (timestamp), is compared with a threshold vector to form a Boolean result, which may have a first value and a second value. If the Boolean result has a first value for the given date (timestamp), then the symbol of the signal object and the date (timestamp) at which that symbol was assigned are transmitted from the sensor to the computer system. If necessary, other parameters can be transmitted based on the identified signal object.

[0390] Particularly advantageous is the reconstruction of the reconstructed ultrasonic echo signal model 610 based on the identified signal objects 122. This ultrasonic echo signal model 610 is generated by linearly overlapping signal waveform models of each of the identified signal objects. This ultrasonic echo signal model is then subtracted from the ultrasonic echo signal 1. This results in a residual signal 660. As a result, signal objects similar to the selected signal objects identified with higher probability are better suppressed. Weaker signal objects appear more readily in the residual signal 660 and can be better identified (see also...). Figure 18 Therefore, preferably, the ultrasonic echo signal model 610 reconstructed from the identified signal object is also subtracted from the ultrasonic echo signal 1 to form a residual signal 660. This residual signal 660 is then used again to form the feature vector signal 138, and the next signal object is identified with a higher probability. Since the initially identified signal object was removed from the input signal according to its weight, it can no longer influence the identification. Therefore, this form of identification provides better results.

[0391] However, this identification method is generally slow. Therefore, it is reasonable to first perform direct first object identification without subtraction while the measurement is still in progress, and then perform a second pattern identification after all sampled values ​​of the ultrasonic echo response of the vehicle environment are detected and after the ultrasonic pulse or ultrasonic pulse train is emitted, in which the ultrasonic echo signal model 610 is subtracted. This takes longer, but is more accurate for this purpose.

[0392] Preferably, if the value of the sampled value of the residual signal 660 is lower than the value of a predetermined threshold curve, then the narrowing classification of the ultrasonic echo signal 1 into a signal object by means of the ultrasonic echo signal model 610 ends.

[0393] Particularly preferably, data transmission is performed in the vehicle via a serial bidirectional single-wire data bus. In this case, the electrical circuit is preferably secured through the vehicle body. Preferably, sensor data is transmitted to the computer system in a current-modulated manner. The computer system then transmits data for controlling the sensor, preferably in a voltage-modulated manner. It has been recognized that the use of PSI5 and / or DSI3 data buses is particularly suitable for data transmission. It has also been recognized that it is particularly advantageous to transmit data to the computer system at a transmission rate > 200 kBit / s and to at least one sensor from the computer system at a transmission rate > 10 kBit / s, preferably 20 kBit / s. It has also been recognized that the transmission of data from the sensor to the computer system should be modulated using a transmit current sent to the data bus, the current intensity of which should be less than 50 mA, preferably less than 5 mA, and preferably less than 2.5 mA. These buses must be matched accordingly for these operating values. However, the basic principles remain valid.

[0394] To perform the above method, a computer system is required, having a data interface with the aforementioned data bus, preferably a single-wire data bus, and supporting the decompression of data compressed in this manner. However, the computer system typically does not perform complete decompression, but rather, for example, only evaluates the timestamp and the type of the identified signal object. The sensor required to perform one of the above methods has at least one transmitter and at least one receiver for generating a received signal; the transmitter and receiver may also exist in combination as one or more transducers. The sensor also has at least: means for processing and compressing the received signal; and a data interface for transmitting data to the computer system via the data bus, preferably a single-wire data bus. For compression, the means for compression preferably has at least one of the following sub-means:

[0395] - Optimal filter;

[0396] - Comparator;

[0397] - A threshold signal generating device for generating one or more threshold signals SW;

[0398] - A differentiator used for finding derivatives;

[0399] - An integrator used to integrate signals;

[0400] - Other filters;

[0401] - Envelope shaper for generating envelope signals (ultrasonic echo signals) from received signals;

[0402] - A correlation filter used to compare the received signal or the signal derived from it with a reference signal.

[0403] In a particularly simple form, the proposed method for transmitting sensor data, especially ultrasonic sensor data, from a sensor to a computer system, particularly a computer system in a vehicle, is implemented as follows:

[0404] The method begins by emitting a train of ultrasonic pulses. Then, ultrasonic signals are received, typically reflected, and a time-discrete received signal is formed, consisting of a time sequence of sampled values. Each sampled value is assigned a time date (timestamp). This time date typically represents the sampling time point. Based on this data stream, a first parameter signal of a first characteristic is determined by means of a first filter based on the sequence of sampled values ​​of the received signal. Here, the parameter signal is preferably again designed as a time-discrete sequence of parameter signal values. Each parameter signal value is assigned exactly one time date (timestamp). Preferably, this date corresponds to the most recent time date of the sampled value used to form the corresponding parameter signal value. Simultaneously, preferably by means of another filter assigned to at least one other parameter signal, other parameter signals of the characteristic assigned to that other parameter signal are determined based on the sequence of sampled values ​​of the received signal, wherein these other parameter signals are each again designed as a time-discrete sequence of other parameter signal values. Here, each other parameter signal value is also assigned the same time date (timestamp) as the corresponding parameter signal value.

[0405] In the following text, the first parameter signal and the other parameter signals are collectively referred to as the eigenvector signal. Therefore, the eigenvector signal (or perhaps the parameter vector signal) is a time-discrete sequence of eigenvector signal values, which are composed of these parameter signal values ​​and other parameter signal values ​​each having the same date (timestamp). Thus, each eigenvector signal value (= parameter signal value) formed in this way can be assigned the corresponding date (timestamp).

[0406] Next, preferably, in the case of forming a Boolean result that can have a first value and a second value, the feature vector signal value of the time and date (timestamp) is compared quasi-continuously with a threshold vector, preferably a prototype vector. For example, it is conceivable to compare the value of the current feature vector signal value, for example, representing the first component of the feature vector signal value, with a fifteenth threshold representing the first component of the threshold vector, and if the value of the feature vector signal value is less than the fifteenth threshold, then the Boolean result is set to the first value, and if not, the Boolean result is set to the second value. If the Boolean result has the first value, then it is further conceivable to compare the value of another feature vector signal value, for example, representing another component of the feature vector signal, with another threshold representing another component of the threshold vector, and if the value of the other feature vector signal value is less than the other threshold, then the Boolean result remains at the first value, and if not, the Boolean result is set to the second value. In this way, all other feature vector signal values ​​can be checked. Of course, other classifiers are also conceivable. Comparisons with multiple different threshold vectors are also possible. That is, these threshold vectors represent prototypes of a pre-given signal shape. These prototypes come from the mentioned library. Preferably, each threshold vector is reassigned a sign.

[0407] Next, in this case, the final step is: if the Boolean result for the time date (timestamp) has a first value, then the symbol, along with, if necessary, the feature vector signal value and the time date (timestamp) assigned to the symbol or the feature vector signal value, is transmitted from the sensor to the computer system.

[0408] Therefore, no other data is transmitted. Furthermore, interference is avoided through multidimensional evaluation.

[0409] Therefore, a sensor system is proposed based on this, comprising: at least one computer system capable of performing one of the methods described above; and at least two sensors, also capable of performing one of the methods described above, such that the at least two sensors can communicate with the computer system through signal object recognition and can also compactly transmit incoming echoes and additionally provide this information to the computer system. Correspondingly, the sensor system is typically configured such that data transmission between the at least two sensors and the computer system is performed according to or can be performed according to the methods described above. That is, within the at least two sensors of the sensor system, each ultrasonic received signal, i.e., at least two ultrasonic received signals, is typically compressed using one of the methods described above and transmitted to the computer system. Here, within the computer system, the at least two ultrasonic received signals are reconstructed into reconstructed ultrasonic received signals. Then, the computer system performs object recognition of objects in the sensor's environment using the reconstructed ultrasonic received signals. That is, unlike prior art, these sensors do not perform this object recognition.

[0410] Preferably, the computer system additionally uses reconstructed ultrasonic received signals and additional signals from other sensors, particularly radar sensors, to perform object recognition of objects in the sensor environment.

[0411] Preferably, as a final step, the computer system creates an environmental map of the sensors or devices that are partly those sensors based on the identified objects.

[0412] Therefore, the proposed ultrasonic sensor system, such as in Figure 7 and 12As exemplarily illustrated, the system preferably includes an ultrasound transducer 100, a feature extractor 111, and estimators 150, 151 or classifiers, as well as a physical interface 101. Preferably, the ultrasound transducer 100 is configured and / or set to receive acoustic ultrasound signals and form an ultrasound transducer signal 102 accordingly. The feature vector extractor 111 is configured and / or set to form a feature vector signal 138 based on the ultrasound transducer signal 102. Preferably, the ultrasound sensor system is configured to: identify signal objects in the ultrasound echo signal 1 by means of estimators 151, 150 and classify these signal objects into signal object categories, where one signal object category may include only one signal object. Here, preferably, each thus identified and classified signal object 122 is assigned at least one assigned signal object parameter and a symbol corresponding to the signal object category assigned to that signal object, or at least one assigned signal object parameter and a symbol for that signal object are determined for each thus identified and classified signal object 122. The symbol of the identified signal object category 122 and at least one assigned signal object parameter of the identified signal object category 122 are transmitted to the upper-level computer system via the data bus.

[0413] Preferably, estimator 150 has a distance determiner 112 and a prototype database 115. Preferably, estimator 150 also has a Viterbi estimator and a signal object database. Estimator 151 may also use a neural network model.

[0414] Therefore, the proposed method for operating the ultrasonic sensor includes corresponding to Figure 11 Steps:

[0415] - Waiting for an ultrasonic receiving signal that is different from the background noise;

[0416] - If the received ultrasound signal is different from the background noise, the system transitions to the state "no signal object prototype" and executes methods for identifying and classifying the signal prototype.

[0417] - If the first signal basic prototype of a pre-given signal basic prototype sequence is identified, then the sequence of states of the signal basic prototype assigned to that sequence is transformed.

[0418] - Follow the basic sequence of signal prototypes until the basic sequence of signal prototypes ends;

[0419] - If the end of the basic prototype sequence of the signal is reached, the existence of a basic prototype of the signal in the basic prototype sequence is inferred, and the signal object assigned to the sequence is inferred, and the signal object is notified by a signal.

[0420] - In the event of a timeout and / or if the basic prototype of the signal is not expected to be present in the sequence, or if the basic prototype of the signal is detected once or multiple times at the next position where it is not expected to be present, the sequence shall be interrupted (without prior notification). Figure 11 (drawn in Chinese)

[0421] - Return to the state "No Signal Object Basic Prototype".

[0422] The aforementioned compression method corresponds to a decompression method, which is preferably used to decompress ultrasonic received data compressed and transmitted in this signal-object-oriented manner in a control computer after receiving the ultrasonic received data from the sensor via a data bus. After receiving new data to be decompressed from the sensor in the control computer, an ultrasonic echo signal model is provided, which initially has no signal ( Figure 15 a). Now, the ultrasonic echo signal model 610 is slowly filled with and approximates the measured signal waveform by successively adding parameterized signal waveforms of the respective prototypes of the signal objects 160 contained in the ultrasonic receiving data, according to the signal objects. Preferably, only signal waveforms of signal objects that satisfy predetermined conditions, such as a predetermined chirp direction, are added. Figure 13 b). Of course, summation can also be performed without any selection criteria. Figure 15 and 16 Based on the summed signal waveform of the prototype signal object identified by the sensor, and based on the ultrasonic echo signal model, a reconstructed ultrasonic received signal is preferably formed in the sensor's control computer. This can be achieved in the form of sampled values ​​of the reconstructed envelope of the received signal (ultrasonic echo signal). Subsequently, for complex methods such as sensor fusion, this reconstructed ultrasonic received signal (reconstructed ultrasonic echo signal) can be used significantly better in the control computer than the assumed obstacle object data transmitted from the sensor to the controller.

[0423] This compressed data transmission between the sensor and computer system via the data bus reduces the data bus load and thus lessens its criticality relative to EMV requirements. On the other hand, it provides idle data bus capacity, for example, during the reception time, for transmitting control commands from the computer system to the sensor and for transmitting status information and other data from the sensor to the computer system. The proposed prioritization ensures that safety-related data is transmitted first, thus avoiding unnecessary dead time for the sensor. Attached Figure Description

[0424] The present invention will then be described in more detail with reference to various embodiments and the accompanying drawings.

[0425] Figure 1 The basic process of signal compression and transmission is shown (see description above).

[0426] Figure 2 The basic process of signal compression and transmission is shown in more detail (see the description above).

[0427] Figure 3 The diagram illustrates a conventional ultrasound echo signal 1 and its routine evaluation, an ultrasound echo signal including the chirped direction, and... Figure 3 The signal object (triangular signal) identified in signal c after discarding unidentified signal components.

[0428] Figure 4 The diagram illustrates the known conventional transmission and the required transmission for compressed data.

[0429] Figure 5 The required transmission of compressed data is shown.

[0430] Figure 6 The required transmission of compressed data is shown.

[0431] Figure 7 The device shown is in the form of an ultrasonic sensor that identifies signal objects, without a data interface, data bus, and controller for simplicity.

[0432] Figure 8 This is used to illustrate the selection of signal basic object prototypes from the prototype database 115 via distance determiner 112.

[0433] Figure 9 The method is used to illustrate the HMM method, which is applied by the Viterbi estimator 113 so as to identify the signal object 122 as the most likely sequence of the signal basic object prototypes based on the sequence of the identified signal basic object prototypes 121.

[0434] Figure 10 The state sequence used to identify a single signal object in the Viterbi estimator 113 is shown.

[0435] Figure 11 A preferred sequence of states is shown in the Viterbi estimator 113 for continuous identification of one of the signal objects typically required for identification in autonomous driving tasks.

[0436] Figure 12 An alternative design scheme with an estimator 151 implementing a neural network model is shown.

[0437] Figure 13This illustrates the decompression of the chirped-down signal component of the transmitted envelope of an ultrasonic signal (ultrasonic echo signal).

[0438] Figure 14 This illustrates the decompression of the chirped-up signal component of the transmitted envelope of an ultrasonic signal (ultrasonic echo signal).

[0439] Figure 15 and 16 The decompression of the transmitted envelope of an ultrasonic signal (ultrasonic echo signal) with chirped upward and chirped downward signal components is shown.

[0440] Figure 17 The original signal (a), the reconstructed signal (b), and their superposition (c) are shown.

[0441] Figure 18 It shows according to Figure 17 The signal waveform also includes a threshold signal, based on which, such as Figure 18 As can be seen in d, it was used to end the decompression.

[0442] Figure 19 An improved device for improved compression is shown.

[0443] Figure 20 An improved alternative device for further enhanced signal data compression is shown.

[0444] Figure 21 An overview of the main components of a sensor system for detecting the vehicle's environment by recognizing obstacle objects, including predictions of changes of these obstacle objects relative to the vehicle during the dead time of the sensor system.

[0445] Figures 22 to 25 Detailed block diagrams are shown to illustrate different methods for data compression (while keeping the corresponding data decompression unchanged), as they may be... Figure 21 The subject of the corresponding block is specified.

[0446] Figure 26 An overview diagram is shown again.

[0447] Figure 27 It shows the use of Figure 21 The block "identifies p obstacle objects" and "predicts a model" or is used for Figure 26 The use of " Figure 24 An example of a block diagram representing a block.

[0448] Figure 28This illustrates the temporal correspondence between the time periods of transmission, oscillation attenuation, and reception in the case of an ultrasonic transducer in a measurement system, and the predicted time periods used to predict changes in identified real obstacle objects between two successive time points based on the measured values ​​and thus the predicted changes. Detailed Implementation

[0449] Figure 3 The time waveform of a conventional ultrasonic echo signal 1 and its routine evaluation are shown in freely chosen units for the coordinate axes. It begins with the transmission pulse train SB, carrying a threshold signal SW. Output 2 is set to logic 1 whenever the envelope signal of the received ultrasonic signal (ultrasonic echo signal 1) exceeds the threshold signal SW. A time-analog interface with digital output levels is involved. Further evaluation is then performed in the sensor's controller. This analog interface, corresponding to existing technology, cannot signal faults or control the sensor.

[0450] Figure 3 b shows the time waveform of the conventional ultrasonic echo signal 1 and its routine evaluation in units of free choice for the coordinate axes, where the amplitude is transmitted along with it. It begins with the transmission pulse train SB, carrying a threshold signal SW. However, whenever the envelope signal of the received ultrasonic signal (ultrasonic echo signal 1) exceeds the threshold signal SW, output 2 is now set to a level corresponding to the amplitude of the detected reflection. A time-analog interface with analog output levels is involved. Further evaluation is then performed in the sensor's controller. This analog interface, corresponding to existing technology, cannot signal faults or control the sensor.

[0451] In order to illustrate, Figure 3 c shows the ultrasonic echo signal, where the chirp direction (e.g., A = chirp-up; B = chirp-down) is marked.

[0452] exist Figure 3 Section d explains the principle of symbolic signal transmission. (Alternative) Figure 3 The signal in c, for example, transmits only two types of triangle objects. Specifically, the first triangle object A is used for the chirping upward case and the second triangle object B is used for the chirping downward case. Simultaneously, the timing and amplitude of the triangle objects are transmitted. If the signal is now reconstructed based on this data, a result is obtained... Figure 3 The signal corresponding to d. Signal components that do not correspond to the triangular signal are removed from this signal. Therefore, unidentified signal components are discarded and significant data compression is performed.

[0453] Figure 4e illustrates a conventional analog transmission of the intersection of the ultrasonic echo signal 1 and the threshold signal SW of the ultrasonic received signal.

[0454] Figure 4 f illustrates the transmission of analyzed data after the ultrasonic echo has been fully received.

[0455] Figure 4 g illustrates the transmission of compressed data, where, in this example, the symbols used for the basic objects of the signal are transmitted as much as possible without compression according to existing techniques.

[0456] Figure 5 The diagram illustrates the transmission of compressed data, where, in this example, the symbols used for basic signal objects are compressed into symbols used for signal objects. First, a first triangular object 59 is identified and transmitted, representing a time series of values ​​exceeding a threshold, followed by a maximum value, and then values ​​below the threshold. Next, a bimodal signal with a saddle point 60 above the threshold is identified. Here, the sequence is characterized by the ultrasonic echo signal 1 exceeding the threshold signal SW, followed by the maximum value of the ultrasonic echo signal 1, then the minimum value above the threshold signal SW, then the maximum value above the threshold signal SW, and then the value below the threshold signal SW. Following this identification, the symbols of the bimodal signal with the saddle point are transmitted. Here, a timestamp is also transmitted. Preferably, other parameters of the bimodal signal with the saddle point, such as the positions of the maximum and minimum values ​​or scaling factors, are also transmitted. Next, a triangular signal is again identified as the basic signal object of the ultrasonic echo signal 1 exceeding the threshold signal SW, followed by the maximum value of the ultrasonic echo signal 1, and then the ultrasonic echo signal 1 below the threshold signal SW. Then, the bimodal signal is identified again, but this time the minimum value of the ultrasonic echo signal 1 is below the threshold signal SW. That is, the bimodal signal can, for example, be processed as a separate signal object if necessary. If easily identifiable, this processing of the signal results in a significant reduction in data.

[0457] Figure 6 It shows the relationship with Figure 3 The corresponding required transmission of compressed data, wherein in this example the symbols for the basic signal object are compressed into symbols for the signal object, wherein not only the envelope signal (ultrasonic echo signal 1) is evaluated but also the confidence signal is evaluated.

[0458] Figure 7A device in the form of an ultrasonic sensor for identifying signal objects is shown, without a data interface, data bus, or controller for simplicity. An ultrasonic transducer 100 is manipulated and measured via a physical interface 101 using an ultrasonic transducer signal 102. The physical interface 101 drives the ultrasonic transducer 100 and processes the ultrasonic transducer signal 102 received from the ultrasonic transducer 100 into an ultrasonic echo signal 1 for subsequent signal object classification. Preferably, the ultrasonic echo signal 1 is a digitized signal with time-spaced sample values. The feature vector extractor 111 has various devices, exemplarily n optimal filters (optimal filter 1 to optimal filter n), where n is a positive integer. The outputs of these optimal filters form an intermediate parameter signal 123. Alternative optimal filters, or those supplementing these optimal filters, may also be used depending on the application.

[0459] - Integrator; and / or

[0460] - Differentiator; and / or

[0461] - Filters; and / or

[0462] - Logarithmizer; and / or

[0463] - FF and DFFT devices; and / or

[0464] - Correlator, and / or

[0465] - Demodulators, which multiply their input signal with a pre-given signal and then filter it; and / or

[0466] - Other signal processing sub-devices; and / or

[0467] - Their combination,

[0468] These devices then generate an n-dimensional intermediate parameter signal 123. The block indicated as "optimal filter" in the figures can only be understood in this respect as a placeholder for such a signal processing block. This signal processing block, indicated as "optimal filter," can also have more than one output, contributing to the n-dimensional intermediate parameter signal 123 multiple times. For example, the subsequent saliency enhancer is used to map the n-dimensional space of the intermediate parameter signal 123 to the m-dimensional space of the eigenvector signal 138. Here, m is a positive integer. Typically, m is less than n. This aspect is used to selectively maximize the parameter values ​​from which each eigenvector signal value of the eigenvector signal 138 is derived. Preferably, this is achieved by means of a linear mapping of the offset values ​​added to the intermediate parameter signal values, determined statistically in the laboratory, and a so-called LDA matrix, with each vector of the respective sampled value of the intermediate parameter signal 123 multiplied by the LDA matrix to obtain the eigenvector signal 138. Preferably, the distance determiner 112 compares each thus obtained eigenvector signal value of the eigenvector signal 138 with each signal basic object prototype in the prototype database 115. To this end, the prototype database 115 contains entries for the centroid vectors of the corresponding signal basic object prototypes in the prototype database 115, such as 141, 142, 143, and 144. Preferably, the distance between the currently examined feature vector signal value of the feature vector signal 138 and the centroid vector just examined in the prototype database 115 is calculated by the distance determiner 112. However, the distance determiner 112 may also calculate, in other ways, an evaluation of the similarity between the centroid vectors of the corresponding signal basic object prototypes in the prototype database 115, such as 141, 142, 143, and 144, and the currently examined feature vector signal value of the feature vector signal 138, in the form of an evaluation value, which is always referred to as distance for simplicity. Distance determiner 112 determines, in such a way, whether the centroid vectors of, for example, 141, 142, 143, and 144 of the signal basic object prototypes in prototype database 115 are sufficiently similar to the current eigenvector signal value of eigenvector signal 138, i.e., whether they have a sufficiently small distance; and if so, which centroid vector of, for example, 141, 142, 143, and 144 of the signal basic object prototypes in prototype database 115 is most similar to the current eigenvector signal value of eigenvector signal 138, i.e., has the smallest distance. If necessary, distance determiner 112 determines a list of centroid vectors of, for example, 141, 142, 143, and 144 of the signal basic object prototypes in prototype database 115, which are sufficiently similar to the current eigenvector signal value of eigenvector signal 138, i.e., have a sufficiently small distance.Preferably, these centroid vectors are sorted by distance and submitted to the Viterbi estimator as a hypothesis list based on their respective distances. The identification results of the distance determiner (also referred to herein as a classifier) ​​are submitted when the individual identified signal basic object 121 is treated as a symbol (e.g., as a prototype database address pointing to the identified signal basic object prototype in prototype database 115) and when the hypothesis list is preferably a list of pairs consisting of the symbol of the identified signal basic object (e.g., a prototype database address pointing to the identified signal basic object prototype in prototype database 115) and the distance to the centroid of the identified signal basic object. The Viterbi estimator then searches for a sequence of signal basic object prototypes that preferably corresponds to a predefined sequence of signal basic object prototypes in its signal object database 116. In this case, for example, consistency is counted as positive and inconsistency as negative, such that an evaluation value is obtained for each entry in signal object database 116 for the time series of the identified signal basic object prototypes. In the case of a hypothesis list, the Viterbi estimator preferably examines all possible paths through the time series traversing the hypothesis list. Preferably, the Viterbi estimator determines its evaluation result while taking into account previously determined distances. This can be achieved, for example, by dividing the values ​​to be added by the distances before addition when forming the evaluation values. In this way, the Viterbi estimator identifies the identified signal objects 122, which are then transmitted via the data bus, if necessary, equipped with appropriate parameters. Importantly, this approach does not involve the identification of objects physically existing in the measurement space of the ultrasound transducer 100 and the transmission of such information, but rather the identification of structures within the ultrasound echo signal 1 and the transmission of such information.

[0469] Figure 8 This is used to illustrate the selection of signal basic object prototypes from the prototype database 115 via distance determiner 112.

[0470] The distance determines 112. Figure 8The orientation of the current eigenvector signal value of eigenvector signal 138, determined in the two-dimensional parameter space of this embodiment, may be very different. It is conceivable that this first eigenvector signal value 146 is too far from the centroid coordinates 141, 142, 143, 144 of the centroid of any of the signal basic object prototypes in the prototype database 115. This distance threshold could, for example, be half of the minimum prototype distance mentioned in the case of the signal basic object prototypes in the prototype database 115. It is also possible that the signal basic object prototypes in the prototype database 115 overlap in the scattering range around their respective centroids 143, 142, and the determined second current eigenvector signal value 145 of eigenvector signal 138 lies within this overlap. In this case, the assumption list may include two signal basic object prototypes with different probabilities due to different distances as additional parameters. Preferably, these probabilities are represented by distance. That is, instead of handing over the (unique) most probable signal basic object to the Viterbi estimator 113, the Viterbi estimator 113 hands over a vector consisting of possible signal basic objects. The Viterbi estimator 113 then searches the time series of these hypothesis lists for possible sequences that have the highest probability, relative to all possible paths through which a pre-given sequence of signal basic objects in its signal object database passes via the hypothesis list received by the Viterbi estimator 113 from the distance determiner 112 and is identified as a possible signal basic object 121. Here, for each hypothesis list, exactly one of the identified signal basic object prototypes from that hypothesis list must traverse that path.

[0471] In the best case, the current eigenvector signal value 148 is within the scattering range (threshold ellipsoid) 147 around the centroid 141 of the unique signal primitive object prototype 141, and the current eigenvector signal value is thus reliably identified by the distance determiner 112 and handed over to the Viterbi estimator 113 as the identified signal primitive object 121.

[0472] It is conceivable that, in order to improve the modeling of the dispersion range of a single signal basic object prototype, the dispersion range could be modeled using multiple signal basic object prototypes, which are circular in shape here, each with its own dispersion range. That is, multiple signal basic object prototypes in prototype database 115 could represent the same signal basic object prototype of the same signal basic object category.

[0473] Preferably, the distance determiner 112 compares each thus obtained feature vector signal value of feature vector signal 138 with each signal basic object prototype of prototype database 115. For this purpose, prototype database 115 contains entries for each signal basic object prototype in prototype database 115, with centroid vectors of, for example, 141, 142, 143, 144 of the corresponding signal basic object prototype in prototype database 115. Preferably, the distance determiner 112 calculates the distance between the currently examined feature vector signal value of feature vector signal 138 and the currently examined centroid vector of prototype database 115. However, the distance determiner 112 may also calculate, in other ways, an evaluation of the similarity between the centroid vectors of, for example, 141, 142, 143, 144 of the corresponding signal basic object prototypes in prototype database 115 and the currently examined feature vector signal value of feature vector signal 138, in the form of an evaluation value, which, for simplicity, is always referred to as distance. Distance determiner 112 determines, in such a way, whether the centroid vectors of the signal basic object prototypes of prototype database 115, such as 141, 142, 143, and 144, are sufficiently similar to the current eigenvector signal value of eigenvector signal 138, i.e., whether they have a sufficiently small distance, and if so, which centroid vector of the signal basic object prototypes of prototype database 115, such as 141, 142, 143, and 144, is most similar to the current eigenvector signal value of eigenvector signal 138, i.e., has the smallest distance. If necessary, distance determiner 112 determines a list of centroid vectors of the signal basic object prototypes of prototype database 115, such as 141, 142, 143, and 144, that are sufficiently similar to the current eigenvector signal value of eigenvector signal 138, i.e., have a sufficiently small distance.

[0474] Figure 9 This is used to illustrate the Hidden Markov Model (HMM) method, which is applied via a Viterbi estimator 113 to identify signal objects 122 as the most probable sequence of signal object prototypes based on the sequence of identified signal object prototypes 121. (The above text refers to...) Figure 9 This was explained.

[0475] Figure 10 The diagram illustrates the state sequence used in the Viterbi estimator 113 to identify a single signal object. (The preceding text refers to...) Figure 10 This was explained.

[0476] Figure 11 This illustrates a preferred state sequence in the Viterbi estimator 113 for continuous identification of one of the signal objects typically required for identification in autonomous driving tasks. (The preceding text refers to...) Figure 11 This was explained.

[0477] Figure 12 An alternative design with estimator 151, implemented as a neural network model, is shown. For simplicity, a data interface, data bus, and controller are omitted. The ultrasound converter 100 is manipulated and measured via a physical interface 101 using an ultrasound converter signal 102. The physical interface 101 drives the ultrasound converter 100 and processes the ultrasound converter signal 102 received from the ultrasound converter 100 into an ultrasound echo signal 1 for subsequent signal object classification. Preferably, the ultrasound echo signal 1 is a digitized signal consisting of a time series of sampled values. The feature vector extractor 111 has various devices, exemplarily n optimal filters (optimal filter 1 to optimal filter n, where n is a positive integer). The outputs of these optimal filters form an intermediate parameter signal 123. Integrators, filters, differentiators, logarithms, and / or other signal processing sub-devices and combinations thereof can be used, depending on the application, to replace or supplement these optimal filters, and these devices then generate the n-dimensional intermediate parameter signal 123. The saliency enhancer, used advantageously, maps the n-dimensional space of the intermediate parameter signal 123 to the m-dimensional space of the eigenvector signal 138. Here, m is a positive integer. Typically, m is less than n. This aspect is used to selectively maximize the parameter values ​​from which each eigenvector signal value of the eigenvector signal 138 is derived. Preferably, this is achieved by using a linear mapping of the offset values ​​added to the intermediate parameter signal values, determined statistically in the laboratory, and the so-called LDA matrix, multiplying the corresponding vectors of the corresponding sampled values ​​of the intermediate parameter signal 123 with the LDA matrix to obtain the corresponding vector signal 138. Next, the estimator 151 attempts to identify signal objects in the data stream of eigenvector signal values ​​of the eigenvector signal 138 using a neural network model 151. Here, each basic signal object prototype and signal object is encoded through the networking of nodes within the neural network model and its parameterization within the neural network model. Then, the estimator 151 outputs the identified signal object 122.

[0478] Figure 13 The decompression of the transmitted envelope signal (ultrasonic echo signal 1) of the ultrasonic received signal, limited to the chirped downward signal, is shown.

[0479] First, a model of an ultrasonic echo signal without a signal is generated. Figure 13 a) The ultrasonic echo signal model is parameterized using model parameters SA, which are related, for example, to the time t since the ultrasonic pulse or ultrasonic pulse train was emitted.

[0480] Next, the ultrasonic echo signal model is preferably supplemented by addition and appropriately parameterized to be transmitted from the sensor to the controller by a first signal object 160 with data transmission priority 1. This first signal object describes a triangular shape with a chirped downward A. Figure 13 b).

[0481] The second signal object 161, which is transmitted with data transmission priority 2 and has a triangular shape with chirp-up characteristics B, is not considered here for the reconstruction of the chirp-down signal.

[0482] The third signal object 162, transmitted with data transmission priority 3 and having a bimodal shape with chirped upward characteristic B, is not considered here for the reconstruction of the chirped downward signal.

[0483] Next, the supplemented ultrasonic echo signal model is preferably supplemented appropriately and parameterized by addition with data transmission priority 4 from the sensor to the controller's fourth signal object 163, which describes a triangular shape with a chirped downward A. Figure 13 c).

[0484] Next, the supplemented ultrasonic echo signal model is preferably supplemented appropriately and parameterized by addition with data transmission priority 5 from the sensor to the controller's fifth signal object 164, which describes a triangular shape with a chirped downward A. Figure 13 d).

[0485] The sixth signal object 165, transmitted with data transmission priority 6 and having a bimodal shape with chirped upward characteristic B, is not considered here for the reconstruction of the chirped downward signal.

[0486] The obtained reconstructed envelope signal is a reconstructed chirped downward ultrasound echo signal.

[0487] The decompressed and reconstructed ultrasonic echo signal is then typically used for object recognition in the controller or elsewhere in the vehicle.

[0488] Figure 14 The decompression of the transmitted envelope signal (ultrasonic echo signal 1) of the ultrasonic received signal, limited to the chirped upward signal, is shown.

[0489] First, a model of an ultrasonic echo signal without a signal is generated. Figure 14 a) The ultrasonic echo signal model is parameterized using model parameters SA, which are related, for example, to the time t since the ultrasonic pulse or ultrasonic pulse train was emitted.

[0490] The first signal object 160, which is a triangle shape with a chirped downward characteristic A and is transmitted with data transmission priority 1, is not considered here for the reconstruction of the chirped upward signal.

[0491] Next, the ultrasonic echo signal model is preferably supplemented with appropriate parameterization by addition, and transmitted from the sensor to the controller via a second signal object 161, which describes a triangular shape with chirped upward B. Figure 14 b).

[0492] Next, the ultrasonic echo signal model is preferably supplemented by a third signal object 161, which is transmitted from the sensor to the controller with data transmission priority 3, by addition in a suitable parameterized manner. This third signal object describes a bimodal shape with chirped upward B. Figure 14 c).

[0493] The fourth signal object 162, which is transmitted with data transmission priority 4 and has a triangular shape with chirped downward characteristic A, is not considered here for the reconstruction of the chirped upward signal.

[0494] The fifth signal object 163, which is transmitted with data priority 5 and has a triangular shape with chirped downward characteristic A, is not considered here for the reconstruction of the chirped upward signal.

[0495] Next, the ultrasonic echo signal model is preferably supplemented by a sixth signal object 165, which is transmitted from the sensor to the controller with a data transmission priority of 6. This sixth signal object describes a triangular shape with a chirped upward B. Figure 14 d).

[0496] The obtained reconstructed envelope signal is a reconstructed chirped upward ultrasound echo signal.

[0497] The decompressed and reconstructed ultrasonic echo signal is then typically used for object recognition in the controller or elsewhere in the vehicle.

[0498] Figure 15 and 16 The decompression of the transmitted envelope signal (ultrasonic echo signal 1) of the ultrasonic received signal with chirped up and chirped down signals is shown.

[0499] First, a model of an ultrasonic echo signal without a signal is generated. Figure 15 a) The ultrasonic echo signal model is parameterized using model parameters SA, which are related, for example, to the time t since the ultrasonic pulse or ultrasonic pulse train was emitted.

[0500] Next, the ultrasonic echo signal model is preferably supplemented by appropriate parameterization through addition, with data transmission priority 1, from the sensor to the controller via a first signal object 160, which describes a triangular shape with a chirped downward A. Figure 15 b).

[0501] Next, the ultrasonic echo signal model is preferably supplemented by appropriate parameterization through addition, with data transmission priority 2, from the sensor to the controller via a second signal object 161, which describes a triangular shape with chirped upward B. Figure 15 c).

[0502] Next, the ultrasonic echo signal model is preferably supplemented by a third signal object 162, which is transmitted from the sensor to the controller with data transmission priority 3, by addition in a suitable parameterized manner. This third signal object describes a bimodal shape with chirped upward B. Figure 15 d).

[0503] Next, the ultrasonic echo signal model is preferably supplemented by appropriate parameterization through addition to a fourth signal object 163 transmitted from the sensor to the controller with data transmission priority 4. This fourth signal object describes a triangular shape with a chirped downward A. Figure 16 e).

[0504] Next, the ultrasonic echo signal model is preferably supplemented by a fifth signal object 164, which is transmitted from the sensor to the controller with a data transmission priority of 5, by addition in a suitable parameterized manner. This fifth signal object describes a triangular shape with a chirped downward A. Figure 16 f).

[0505] Next, the ultrasonic echo signal model is preferably supplemented by a sixth signal object 165, which is transmitted from the sensor to the controller with a data transmission priority of 6. This sixth signal object describes a triangular shape with a chirped upward B. Figure 16 g).

[0506] To clarify, in Figure 16 The reconstructed envelope signal is drawn in bold in h. In this respect, Figure 16 h and Figure 16 g is consistent.

[0507] The obtained reconstructed envelope signal is a reconstructed chirped up / chirped down ultrasonic echo signal.

[0508] The decompressed and reconstructed ultrasonic echo signal is then typically used for object recognition in the controller or elsewhere in the vehicle.

[0509] Figure 17 The original signal (a), the reconstructed signal (b), and their superposition (c) are shown. The deviation between the reconstructed ultrasound echo signal 166 and the original ultrasound echo signal is very small when the procedure is correct and the signal base and signal object are carefully selected.

[0510] exist Figure 18 In addition to Figure 17 In addition to the three charts, the threshold signal 670 is also shown. Based on... Figure 18 As shown in Figure d, since the residual signal 660 is less than the threshold signal 670, the process of extracting features from the echo signal ends.

[0511] Figure 19 An apparatus for improved compression is shown. Figures 13 to 17 The reconstructor 600, as described in its functional description, can be used not only in the controller but also in the sensor itself before data transmission. For this purpose, the reconstructor 600 generates a reconstructed ultrasonic echo signal model 610, as shown in... Figures 13 to 17 As indicated by the purpose, the reconstructed ultrasonic echo signal model is subtracted from the previously received ultrasonic echo signal 1 in subtractor 602. For this purpose, ultrasonic echo signal 1 is preferably stored in memory 601 in the form of digital sampled values, preferably ordered by time. Here, the stored sampled values ​​of ultrasonic echo signal 1 preferably correspond to exactly one value of the reconstructed ultrasonic echo signal model 610, which is also preferably stored in reconstruction memory 603. Reconstruction memory 603 can be part of reconstructor 600. The additional feedback branches 600, 610, 603, 602, and 601 can also be used for other classifiers, such as those used for... Figure 12 The device.

[0512] exist Figure 20 The text shows the relationship with... Figure 18 Similar devices, but Figure 19 The arrangement of the blocks Figure 20 The text has been changed.

[0513] Figure 21An overview of a single component of a system for identifying obstacle objects in an environment such as a vehicle is shown. The system first includes data compression of sensor signal data from sensors S1, S2, and Sm based on signal waveform characteristics, wherein this data compression is performed without checking or being able to check which actual obstacle objects are present based on the sensor signals. The compressed sensor signal data is transmitted via a data bus to a data processing unit, where the identification of actual obstacle objects and the prediction of changes in these obstacle objects are performed during multiple prediction time periods. The results of the inspection can be signaled by display, more specifically, optically and / or acoustically and / or by displaying pictographs symbolically representing the identified obstacle objects, etc., and / or tactilely.

[0514] in accordance with Figure 22 The process of compressing the data and transmitting it to the data processing unit (ECU) is then described.

[0515] Figure 22 The diagram illustrates the reception and compression of ultrasonic signals in an ultrasonic sensor having a driver DK for a US converter TR and a receiver RX, the transmission of compressed data to a data processing unit ECU via a data bus DB, and the decompression of the compressed data in the data processing unit ECU.

[0516] The following describes the reception and signal processing of ultrasonic signals in an ultrasonic sensor.

[0517] The ultrasound transducer TR is controlled by a driver DR, which receives control signals from a data processing unit ECU for this purpose. A receiver RX, connected to both the ultrasound transducer TR and the driver DR, processes the received ultrasound signals, enabling feature extraction FE, also connected to the receiver RX, to be performed. The result of feature extraction FE is a total feature vector signal F0, which is stored in a zeroth buffer memory IM0. This zeroth buffer memory IM0 transfers the total feature vector signal F1 to a zeroth summer A0, which then feeds the total feature vector signal F1 to an artificial neural signal object recognition network NN0. This network is trained to identify individual signal objects from the total feature vector signal F1.

[0518] The ultrasonic sensor also includes a first buffer memory IM1, a second buffer memory IM2, and others. Figure 1The buffers, not shown, are numbered in ascending order up to the nth buffer IMn. In this context and hereinafter, n is a natural number greater than or equal to 2 and represents the number of storable, identifiable signal objects. The zeroth buffer IM0 up to the nth buffer IMn is controlled by an identification control RC, which in turn is controlled by a system control SCU.

[0519] The first buffer memory IM1 stores the identified first signal object O1. The second buffer memory IM2 stores the identified second signal object O2. The nth buffer memory IMn stores the identified nth signal object On. The identified first signal object O1 is transferred to the first artificial neural single reconstruction network NN1 as input value from the first buffer memory IM1. The identified second signal object O2 is transferred to the second artificial neural single reconstruction network NN2 as input value from the second buffer memory IM2. The identified nth signal object On is transferred to the nth artificial neural single reconstruction network NNn as input value from the nth buffer memory IMn.

[0520] The first neural single-reconstruction network NN1 reconstructs the first single feature vector signal R1 based on the identified first signal object O1. The second neural single-reconstruction network NN2 reconstructs the second single feature vector signal R2 based on the identified second signal object O2. The nth neural single-reconstruction network NNn reconstructs the nth single feature vector signal Rn based on the identified nth signal object On.

[0521] The reconstructed individual eigenvector signals R1, R2, ..., Rn are summed to form the reconstructed total eigenvector signal RF. To do this, the nth reconstructed individual eigenvector signal Rn is added at the nth summer An to the (n-1)th reconstructed individual eigenvector signal Rn-1. Correspondingly, each i-th reconstructed individual eigenvector signal is added at the i-th summer An to the (i-1)th reconstructed eigenvector signal, where i is any natural number between 1 and n. Therefore, the second reconstructed individual eigenvector signal R2 is added at the second summer A2, and the first reconstructed individual eigenvector signal R1 is added at the first summer A1. The sum of all the reconstructed individual eigenvector signals results in the reconstructed eigenvector signal RF.

[0522] At the zeroth summer A0, the reconstructed feature vector signal RF is subtracted from the total feature vector signal F1. The result is the feature vector residual signal F2, which is then passed as input to the neural signal object recognition network NN0 by the zeroth summer A0.

[0523] The neural signal object recognition network NN0 identifies objects from the residual signal F2 of the feature vector. The first identified signal object O1 is transferred to the first buffer memory IM1. The second identified signal object O2 is transferred to the second buffer memory IM2, and so on, until the last identified signal object On is transferred to the nth buffer memory IMn.

[0524] The process of subtracting the reconstructed feature vector signal RF from the total feature vector signal F1, and then performing object recognition and reconstruction of the total feature vector signal F1 using individual neural reconstruction networks NN1, NN2, ..., NNn, is continued until the reconstructed feature vector signal RF is identical to the total feature vector signal F1, that is, until the residual feature vector signal F2 equals zero. This means that the signal objects in the total feature vector signal F1 were correctly identified by the neural signal object recognition network NN0.

[0525] The first identified signal object O1 is transferred from the first buffer memory IM1 to the ultrasonic transmitter control device TU. The second identified signal object O2 is transferred from the second buffer memory IM2 to the ultrasonic transmitter control device TU. The nth identified signal object On is transferred from the nth buffer memory IMn to the ultrasonic transmitter control device TU. The ultrasonic transmitter control device TU then transfers the compressed, identified signal objects to the data bus interface TRU of the data bus DB, and transmits the compressed data to the data bus interface TRE of the data processing unit ECU via the data bus.

[0526] The identification control RC and the ultrasonic transmitter control unit TU are controlled by the ultrasonic system control SCU via the internal data bus IDB.

[0527] Subsequently, according to Figure 22 The right side will be used to explain the decompression of data within the data processing unit (ECU).

[0528] The data bus interface TRE of the data processing unit ECU transfers the data received via the data bus DB to the controller TEC of the data processing unit ECU.

[0529] The controller TEC of the data processing unit ECU transfers the data received via the data bus interface TRE, representing the first transmitted signal object E1, to the first buffer memory EM1, and transfers the data received via the data bus interface TRE, representing the second transmitted signal object E2, to the second buffer memory EM2 of the data processing unit ECU. Correspondingly, the received data representing the i-th transmitted signal object is transferred to the i-th buffer memory of the data processing unit ECU, where i is a natural number between 2 and n. Here, n is also the number of storable and identified signal objects. Finally, the received data representing the n-th transmitted signal object En is transferred to the n-th buffer memory EMn of the data processing unit ECU.

[0530] The first transmitted signal object E1 is transferred to the first neural decompression network ENN1 of the data processing unit ECU as input value via the first buffer memory EM1. The second transmitted signal object E2 is transferred to the second neural decompression network ENN2 of the data processing unit ECU as input value via the second buffer memory EM2. The nth transmitted signal object En is transferred to the nth neural decompression network ENNn of the data processing unit ECU as input value via the nth buffer memory EMn.

[0531] The neural single-reconstruction networks ENN1, ENN2, ..., ENNn of the data processing unit ECU are parameterized identically to their corresponding neural single-reconstruction networks NN1, NN2, ..., NNn in the ultrasound sensor. The first neural single-decompression network ENN1 of the data processing unit ECU is parameterized identically to the first neural single-reconstruction network NN1 in the ultrasound sensor. The second neural single-decompression network ENN2 of the data processing unit ECU is parameterized identically to the second neural single-reconstruction network NN2. Correspondingly, the i-th neural single-decompression network ENNn of the data processing unit ECU is parameterized identically to the i-th neural single-reconstruction network NNn in the ultrasound sensor, where i is a natural number between 1 and n. Finally, the n-th neural single-decompression network ENNn of the data processing unit ECU is parameterized identically to the n-th neural single-reconstruction network NNn. Here, n also corresponds to the number of storable, identifiable objects.

[0532] The first neural single decompression network ENN1 of the data processing unit ECU reconstructs the reconstructed first single feature vector signal ER1 based on the data transmitted for the transmitted first signal object E1. The second neural single decompression network ENN2 of the data processing unit ECU reconstructs the reconstructed second single feature vector signal ER2 based on the data transmitted for the transmitted second signal object E2. The nth neural single decompression network ENNn of the data processing unit ECU reconstructs the reconstructed nth single feature vector signal ERn based on the data transmitted for the transmitted nth signal object En.

[0533] The reconstructed first single feature vector signal ER1 in the data processing unit ECU, the reconstructed second single feature vector signal ER2 in the data processing unit ECU, and all other reconstructed single feature vector signals in the data processing unit ECU up to the reconstructed nth single feature vector signal ERn are summed to form the reconstructed total feature vector signal ER in the data processing unit ECU.

[0534] Therefore, the reconstructed nth individual eigenvector signal ERn is added to the reconstructed (n-1)th individual eigenvector signal ERn-1 at the nth summer ESn of the data processing unit ECU. Correspondingly, each reconstructed ith individual eigenvector signal is added to the reconstructed (i-1)th individual eigenvector signal at the i-th summer of the data processing unit ECU, where i is any natural number between 1 and n. Thus, the reconstructed second individual eigenvector signal ER2 is added at the second summer ES2 of the data processing unit ECU, and the reconstructed first individual eigenvector signal ER1 is added at the first summer ES1 of the data processing unit ECU. The sum of all the reconstructed individual eigenvector signals results in the reconstructed total eigenvector signal ER of the data processing unit ECU.

[0535] exist Figure 23 The text describes alternatives to the compression method, which have been previously discussed. Figure 22The left side is described. In this variant, the sensor signal from the receiver RX of the US sensor is stored in a buffer memory IM0. After the signal object is extracted using the neural signal object recognition network NN0 as described above, the signal object is stored in the buffer memory set up for this purpose, and individual feature vectors R1, R2, ..., Rn are generated by the individual neural reconstruction networks NN1, NN2, ..., NNn assigned to this buffer memory. Now, the relevant individual feature vector signals are inversely transformed into the time domain, more precisely by means of the so-called "inverse" feature extraction IFE, which results in a representation of the signal object in the time domain, i.e., a curve waveform. Now, this curve waveform is subtracted from the sensor signal to form a residual sensor signal, which is fed to the signal object recognition network as a residual total feature vector signal after feature extraction FE. Then, the above process is repeated accordingly.

[0536] The iterative process described above for the continuous and individual extraction of signal objects can also be correspondingly applied to the above-mentioned basis. Figure 22 This is achieved through the described data compression process. In this case, individual eigenvector signals are progressively subtracted from the total eigenvector signal. This can also be achieved by using only a single, unique neural reconstruction network, rather than allocating a separate single reconstruction network to each buffer.

[0537] exist Figure 23 In this context, REK is used to denote which components belong to the so-called reconstructor, such as in... Figure 19 The figure is denoted by reference numeral 600. That is, in this respect, Figure 23 The dashed block REK in the middle corresponds to according to Figure 19 The rebuilder 600. The rebuild memory RSP corresponds to Figure 19 Reconstruction memory 603.

[0538] in accordance with Figure 24 The text then describes another variation of data compression, as previously discussed. Figure 22 and 23 The left half refers to the methods described elsewhere. Here, as in conjunction with... Figure 23 As mentioned above, an inverse time-domain transform is also performed. However, unlike in accordance with... Figure 23 In the inverse transformation method, each individual feature vector signal R1, R2, Rn-1, Rn is inversely transformed to the time domain in its own inverse feature extraction IFE1, IFE2, ..., IFEn-1 and stored in the reconstruction memory RSP respectively, so as to be gradually accumulated there as needed, more precisely, gradually accumulated after each subtraction from the sensor signal and identification of potential other signal objects in the residual sensor signal.

[0539] according to Figure 23 The REK reconstructor can also be used to replace the sequential extraction or identification of signal objects in a serial manner, allowing the process to be performed in parallel. Figure 25 This is illustrated in the diagram. The neural signal object recognition network NN0 is connected to the individual buffers IM1, IM2, ..., IMn via separate lines. The individual feature vector signals R1, R2, Rn-1, Rn generated by the individual neural reconstruction networks NN1, NN2, ..., NNn are inversely transformed back to the time domain using their own inverse feature extraction methods IFE1, IFE2, IFEn-1, IFEn. The sum of all inversely transformed signal objects (i.e., the corresponding curve waveforms) is calculated. The result can be stored in the reconstruction memory RSP, but it is not mandatory.

[0540] According to Figure 25 The parallel processing possibility for signal object extraction shown in the embodiments can also be similarly applied to... Figures 22 to 24 This is implemented in the example. Next, the neural signal object recognition network NN0 is connected to the buffer memories IM1, IM2, ..., IMn via separate lines.

[0541] If you want to perform signal object recognition step by step, that is, sequentially, as based on Figures 22 to 24 As shown, the allocation of sequentially determined signal objects to the buffer memory should also be controlled. This ensures that only one signal object is stored in the buffer memory, and that the signal object is not overwritten as the process progresses.

[0542] in accordance with Figure 26 and 27 The recognition of obstacle objects is then described based on the reconstructed total feature vector signals EF1, EF2, ..., EFm from multiple sensors S1, S2, S3, ..., Sm, where... Figure 27 The model is also shown in the form of a block diagram. Figure 27 (The right side of the model) This model can be used to predict changes in the relative position of an obstacle object relative to the vehicle due to the relative motion of the two.

[0543] Figure 26The diagram schematically illustrates multiple sensors S1, S2, S3, ..., Sm, whose compressed sensor signal data is transmitted to a data processing unit (ECU). This ECU may be the same as or a separate data processing unit. Reconstruction is assigned to the transmitted compressed sensor signal data for each sensor, and decompression of this data is performed (see SR1ER, SR2ER, SR3ER, ..., SRmER). The corresponding reconstructed total characteristic vector signals EF1, EF2, ..., EFm are then fed to the user. Figure 27 The function block represents a virtual reality vector signal whose output is a virtual reality vector signal that depicts the vehicle's environment for a given point in time, for example, in the form of an environment map.

[0544] Figure 27 A block diagram is shown, which includes the reality portion on the left and the model portion on the right. Figure 27 The input parameters are multiple reconstructed total characteristic vector signals ER from the data processing unit (ECU) for m different sensors S1, S2, S3, ..., Sm or m different sensor systems. To avoid [interference with...] Figure 22 The reference numerals ER1 to Ern are confused, and the reference numerals EF1, EF2 to EFm are used to represent the total feature vector signals reconstructed from m sensors S1 to Sm at different time points.

[0545] Within the reality section, the reconstructed total characteristic vector signals EF1, EF2, to EFm, representing the signals from different sensors S1, S2, S3, ..., Sm (hereinafter referred to as the total characteristic vector signals of the sensors), are processed to identify obstacle objects HO1, HO2, ..., HOp. In the model section, based on the obstacle objects HO1, HO2, ..., HOp identified in the reality section, the spatial positions of the obstacle objects between measurement points in time are predicted; these spatial positions can be read at a higher sampling rate. The problem to be solved is that, although the time interval between ultrasonic measurements emitting temporally successive ultrasonic pulses (pulse trains) is not arbitrarily short, conclusions should still be drawn regarding the changes in the obstacle object's orientation relative to the vehicle. That is, the model section quasi-simulates reality so that the vehicle can navigate between ultrasonic measurements during autonomous driving, i.e., without emitting ultrasonic pulses at specific time points. The actual measurement rate should be chosen to be as high as possible so that the model can be compared with the actual measurements between each measurement, thus making the model as close to reality as possible. This comparison is performed through cooperation between the reality section and the model section. Figure 27The block diagram shown can preferably be implemented in a vehicle's data processing unit (ECU), such as an ultrasonic measurement system controller.

[0546] First, a block diagram of the reality part is described, namely... Figure 27 The right side of.

[0547] In the practical part, the method is performed periodically during the identification period EP, which is divided into periods (see [link to relevant documentation]). Figure 28 ).

[0548] At the first summer RS1, the reconstructed first feature vector signal EV1 of the real portion is subtracted from the reconstructed total feature vector signal EF of the sensor signal for the first sensor S1 (which is subsequently represented by the total feature vector signal of the sensor). Here, the term "sensor" is used as a more compact expression. However, those skilled in the art will recognize that a sensor system can also behave like a sensor, so the name "sensor" always refers to "sensor system." The result of the subtraction is the modified first feature vector signal EC1, which serves as the input value for the neural total obstacle recognition network ANN0. That is, the result of this subtraction is simply passed to the artificial neural total obstacle recognition network ANN0, which differs from the existing obstacle object recognition results in the form of the reconstructed first feature vector signal EV1. Therefore, at the beginning of the recognition period, the value of the reconstructed first feature vector signal EV1 is set to zero, so that this value does not change during this phase.

[0549] At the second summer RS2, the reconstructed second feature vector signal EV2 is subtracted from the reconstructed total feature vector signal EF2 from the second sensor S2. The result of the subtraction is a modified second feature vector signal EC2, which serves as the input value for the neural obstacle recognition network ANN0. That is, the result of this subtraction is simply passed to the neural obstacle recognition network ANN0, which differs from existing obstacle object recognition results in the form of the reconstructed second feature vector signal EV2. Therefore, at the beginning of the recognition period, the value of the reconstructed second feature vector signal EV2 is set to zero, so that this value does not change during this phase.

[0550] At the third summer RS3, the reconstructed third feature vector signal EV3 is subtracted from the reconstructed total feature vector signal EF3 from the third sensor S3. The result of the subtraction is a modified third feature vector signal EC3, which serves as the input value for the neural obstacle recognition network ANN0. That is, only this subtraction result is passed to the neural obstacle recognition network ANN0, which differs from the existing obstacle object recognition results in the form of the reconstructed third feature vector signal EV3. Therefore, at the beginning of the recognition period, the value of the reconstructed third feature vector signal EV3 is set to zero, so that the value does not change during this phase.

[0551] At the m-th summer RSm, the reconstructed m-th feature vector signal EVm is subtracted from the reconstructed total feature vector signal EFm from the m-th sensor Sm. The result of the subtraction is the modified m-th feature vector signal ECm, which serves as the input value for the neural obstacle recognition network ANN0. In this case, the variable m represents the number of sensors. That is, this result is simply passed to the neural obstacle recognition network ANN0, which differs from the existing recognition results in the form of the reconstructed m-th feature vector signal EVm. Therefore, at the beginning of the recognition period, the value of the reconstructed first feature vector signal EV1 is set to zero, so that this value does not change during this phase.

[0552] The neural obstacle recognition network ANN0 identifies the most likely and dominant obstacle objects HO1, HO2, ..., HOp from the modified feature vector signals EC1 to ECm within one cycle.

[0553] During the first cycle, the first obstacle object HO1 identified by the overall obstacle recognition network ANN0 is stored in the first buffer memory EO1 of the reality portion. As a result, the reconstructed first feature vector signal EV1 changes, which will be explained later.

[0554] During the second cycle, the second obstacle object H02, identified by the overall obstacle recognition network ANN0, is stored in the second buffer memory E02 of the reality portion. Consequently, the reconstructed second feature vector signal EV2 changes, as will be explained later.

[0555] During the p-th cycle, the p-th obstacle object HOP identified by the total obstacle recognition network ANN0 is stored in the p-th buffer memory EOp of the reality part.

[0556] In this case, the variable p represents the number of obstacle objects identified and thus the number of cycles for each identification period.

[0557] The first neural single obstacle object recognition network RNN1 in the real part reconstructs m feature vector signals RO11 to RO1m from information stored in the first buffer memory EO1. These feature vector signals at least partially represent the identified first obstacle object HO1 (hereinafter referred to as the feature vector signal of the identified first obstacle object). Moreover, these feature vector signals, together with other feature vector signals RO22, ..., ROp1, are used to correct the input of the total feature vector signals EF1, EF2, ..., EFm of the sensors S1, S2, ..., Sm, which is described further below. The m reconstructed feature vector signals RO11 to RO1m representing the first obstacle object HO1 include: RO11, a reconstructed feature vector signal of the first obstacle object HO1 used to correct the total feature vector signal EF1 of the first sensor S1; RO12, a reconstructed feature vector signal of the first obstacle object HO1 used to correct the total feature vector signal EF2 of the second sensor S2; RO13, a reconstructed feature vector signal of the first obstacle object HO1 used to correct the total feature vector signal EF3 of the third sensor S3, and so on, up to RO1m, a reconstructed feature vector signal of the first obstacle object HO1 used to correct the total feature vector signal EFm of the m-th sensor Sm. Furthermore, the first neural single obstacle object recognition network RNN1 outputs the reconstructed feature vector signal RO1 of the first obstacle object HO1 for all m sensors S1 to Sm. The reconstructed feature vector signal RO1 for the first obstacle object HO1 with all m sensors S1 to Sm represents a modeling conception of the obstacle object having identified characteristics such as spatial position, orientation, object type, direction of movement, object speed, etc. Preferably, this is at least in part a cascade of other reconstructed feature vector signals RO11 to RO1m for the first obstacle object HO1 with all m sensors S1 to Sm in a common column vector. If a vector is referred to here, it means a signal with values ​​that are represented by signals in a time- or spatial multiplexed manner, where these values ​​then form tuples representing vectors. At the end of the first cycle, the reconstructed feature vector signal RO1 for the first obstacle object HO1 with all m sensors S1 to Sm is temporarily stored in the first obstacle object memory SP1. The output of the first obstacle object memory SP1 is the reconstructed and stored first feature vector signal ROS1 for the first obstacle object HO1.

[0558] The second neural single obstacle object recognition network RNN2 in the reality part reconstructs m feature vector signals RO21 to RO2m from information stored in the second buffer memory EO2. These feature vector signals at least partially represent the second obstacle object HO2 (hereinafter referred to as the feature vector signal of the recognized second obstacle object), and these feature vector signals, together with other feature vector signals RO21, ..., ROp, are used to correct the input of the total feature vector signals EF1, EF2, ..., EFm of the sensors S1, S2, ..., Sm, which is described further below. The representation of the second obstacle object HO2 by the m reconstructed feature vector signals RO21 to RO2m includes: the reconstructed feature vector signal RO21 of the second obstacle object HO2, used to correct the total feature vector signal EF1 of the first sensor S1; the reconstructed feature vector signal RO22 of the second obstacle object H02, used to correct the total feature vector signal EF2 of the second sensor S2; the reconstructed feature vector signal RO23 of the second obstacle object HO2, used to correct the total feature vector signal EF3 of the third sensor S3, and so on, up to the reconstructed feature vector signal RO2m of the second obstacle object HO2 used to correct the total feature vector signal EFm of the m-th sensor Sm. Furthermore, the second neural single obstacle object recognition network RNN2 outputs the reconstructed feature vector signal RO2 of the second obstacle object HO2 for all m sensors S1 to Sm. The reconstructed feature vector signal RO2 for the second obstacle object HO2 for all m sensors S1 to Sm represents a modeling conception of the obstacle object having identified characteristics such as spatial position, orientation, object type, direction of movement, object speed, etc. Preferably, this is at least in part a cascade of other reconstructed feature vector signals R21 to R2m for the second obstacle object HO2 for all m sensors S1 to Sm in a common column vector. If a vector is referred to here, it means a signal with values ​​that are represented by signals using temporal or spatial multiplexing, where these values ​​thus form tuples representing vectors. At the end of the second cycle, the reconstructed feature vector signal RO2 for the second obstacle object HO2 for all m sensors S1 to Sm is temporarily stored in the second obstacle object memory SP2. The output of the second obstacle object memory SP2 is the reconstructed and stored second feature vector signal ROS2 for the second obstacle object HO2.

[0559] The p-th neural single obstacle object recognition network RNNp in the reality part reconstructs m feature vector signals ROp1 to ROpmm from information stored in the p-th buffer E02. These feature vector signals at least partially represent the first identified obstacle object HO1 (hereinafter referred to as the feature vector signal of the first identified obstacle object), and these feature vector signals, together with other feature vector signals ROp1, are used to correct the input of the total feature vector signals EF1, EF2, ..., Sm of the sensors S1, S2, ..., Sm, which is described further below. The m reconstructed feature vector signals ROp1 to ROpm representing the first obstacle object H=1 include: the reconstructed feature vector signal ROp1 of the p-th obstacle object HOP, used to correct the total feature vector signal EF1 of the first sensor S1; and the reconstructed feature vector signal ROp2 of the p-th obstacle object HOP, used to correct the total feature vector signal of the second sensor S2; and the reconstructed feature vector signal ROp3 of the p-th obstacle object HOP, used to correct the total feature vector signal EF3 of the third sensor S3, and so on, up to the reconstructed feature vector signal ROpm of the p-th obstacle object HOP used to correct the total feature vector signal EFm of the m-th sensor Sm. Furthermore, the p-th neural single obstacle object recognition network RNNp outputs the reconstructed feature vector signal ROp of the p-th obstacle object HOP for all m sensors S1 to Sm. The reconstructed feature vector signal ROp for the p-th obstacle object HOp for all m sensors S1 to Sm represents a modeling conception of an obstacle object with identified characteristics such as spatial position, orientation, object type, direction of movement, object speed, etc. Preferably, this is at least in part a concatenation of other reconstructed feature vector signals Rp1 to Rpm for the p-th obstacle object HOp for all m sensors S1 to Sm in a common column vector. If a vector is referred to here, it means a signal with values ​​that are represented by signals in a time- or spatial multiplexed manner, where these values ​​thus form tuples representing vectors. At the end of the p-th cycle, the reconstructed feature vector signal ROp for the p-th obstacle object HOp for all m sensors S1 to Sm is temporarily stored in the p-th obstacle object memory SPp. The output of the p-th obstacle object memory SPp is the reconstructed and stored p-th feature vector signal ROSp of the p-th obstacle object HOp.

[0560] The reconstructed first feature vector signal EV1 is the sum of the reconstructed feature vector signal RO11 for the first obstacle object HO1 of the first sensor S1, the reconstructed feature vector signal RO21 for the second obstacle object HO2 of the first sensor S1, and all other reconstructed feature vector signals RO31 to ROp1 for the first sensor S1 up to the reconstructed feature vector signal ROp1 for the p-th obstacle object HOp of the first sensor S1.

[0561] The reconstructed second feature vector signal EV2 is the sum of the reconstructed feature vector signal RO12 for the first obstacle object HO1 of the second sensor S2, the reconstructed feature vector signal RO22 for the second obstacle object HO2 of the second sensor S2, and all other reconstructed feature vector signals RO32 to ROp2 for the second sensor S2 up to the reconstructed feature vector signal Op2 for the p-th obstacle object HOp of the second sensor S2.

[0562] The reconstructed third feature vector signal EV3 is the sum of the reconstructed feature vector signal RO13 for the first obstacle object HO1 of the third sensor S3, the reconstructed feature vector signal R023 for the second obstacle object HO2 of the third sensor S3, and the reconstructed feature vector signal ROp3 for the p-th obstacle object HOp of the third sensor S3, and all other reconstructed feature vector signals RO33 to ROp3 for the third sensor S3.

[0563] The reconstructed m-th feature vector signal EVm is the sum of the reconstructed feature vector signal RO1m for the first obstacle object HO1 of the m-th sensor Sm, the reconstructed feature vector signal RO2m for the second obstacle object HO2 of the m-th sensor Sm, and the reconstructed feature vector signal ROpm for the p-th obstacle object HOp of the m-th sensor Sm, up to the sum of all other reconstructed feature vector signals RO3m to ROpm for the m-th sensor Sm.

[0564] At the end of the recognition period, i.e. after p cycles, the obstacle object memory SP1 to SPp contains the reconstructed feature vector signals ROS1 to ROSp for the first to pth obstacle objects HO1, HO2, ..., HOp.

[0565] These obstacle object memories, SP1 to SPp, can be read to display, for example, real-world obstacle objects in an environment map. The model described below is based on these contents of the obstacle object memories SP1 to SPp to form predictions of changes in obstacle objects over a period of time up to the next point in time where a result (change in obstacle object) exists based on a measurement.

[0566] The obstacle object memories SP1 to SPp are controlled by the memory control system SPC. The memory control system SPC, as described in the following sections, sets the output vectors, that is, the reconstructed feature vector signals ROS1 to ROSp, to zero at specific predicted time points.

[0567] Now, based on the following text Figure 28 To describe the model part.

[0568] Now, the model part predicts the next development of the reconstructed feature vector signals ROS1 to ROSp of the first to p-th obstacle objects HO1, HO2, ..., HOp based on the current recognition results and previous recognition results.

[0569] In this scenario, the memory-controlled SPC sets the output vectors—that is, the reconstructed eigenvector signals ROS1 to ROSp—to zero during prediction, allowing the model part to operate independently of the reality part. Only when new measurements exist in the reality part after a retransmission of the ultrasound pulse, and thus new reconstructed eigenvector signals ROS1 to ROSp exist, does the memory-controlled SPC convert the output vectors of the object memories SP1 to SPp to these reconstructed eigenvector signals ROS1 to ROSp. In this way, the simulation part obtains updated values ​​from reality, and the prediction can be improved.

[0570] Predictions are made during prediction periods PP1, PP2, ..., PPq, which are further divided into prediction cycles PZ1, PZ2, ..., PZp. The number of these prediction cycles equals the number of obstacle objects HO1, HO2, ..., HOp. To better illustrate this, in... Figure 28 This is shown in (see) Figure 28 (Description).

[0571] The oscillator OSC generates a time base vector TBV (neither is shown in the accompanying figures). The time base vector TBV is distributed to the total neural obstacle object recognition network ANN0 and the individual neural reconstruction networks NN1 to NNp in the reality part, as well as to the total prediction network DANN0 and multiple individual prediction networks MNN1, ..., MNNp. The time base vector TBV enables time-based control and updating of predictions.

[0572] The neural overall prediction network DANN0 is parameterized identically to the neural overall obstacle object recognition network ANN0. This neural overall prediction network, together with other individual artificial neural prediction networks MNN1, ..., MNNp, is used to predict the changes of obstacle objects HO1, HO2, ..., HOp during the prediction time period. The changes of the obstacle objects are then referred to as virtual obstacle objects or virtually recognized obstacle objects.

[0573] At the first summer MS1 in the model section, the reconstructed second feature vector signal RV12 of the first virtual obstacle object HO1 is subtracted from the reconstructed feature vector signal ROS1 of the first obstacle object HO1. The result of the subtraction is the input value in the form of the first prediction feature vector signal PV1 for the neural total prediction network DANN0. At the beginning of the prediction period, the reconstructed second feature vector signal RV12 of the first virtual obstacle object HO1 has a value of 0 and is reset to zero at that time point if necessary.

[0574] At the second summer MS2 in the model section, the reconstructed second feature vector signal RV22 of the second virtual obstacle object HO2 is subtracted from the reconstructed feature vector signal ROS2 of the second obstacle object HO2. The result of the subtraction is the input value in the form of the second prediction feature vector signal PV2 for the neural total prediction network DANN0. At the beginning of the prediction period, the reconstructed second feature vector signal RV22 of the second virtual obstacle object H02 has a value of 0 and is reset to zero at that time point if necessary.

[0575] A similar approach is taken with respect to the feature vector signals of the third obstacle object HO3 up to the (p-1)th obstacle object Hop (including the (p-1)th obstacle object HOp). However, for clarity, it is not included in... Figure 27 This is shown in the text.

[0576] Finally, at the p-th summer MSp in the model section, the reconstructed second feature vector signal RVp2 of the p-th virtual obstacle object HOP is subtracted from the reconstructed and stored feature vector signal ROSp of the p-th obstacle object HOP. The result of this subtraction is an input value in the form of the p-th predicted feature vector signal PVp for the neural total prediction network DANN0. At the start of the prediction period, the reconstructed second feature vector signal RVp2 of the p-th virtual obstacle object HOP has a value of 0 and is reset to zero at that time point if necessary.

[0577] In addition to predicting the feature vector signals PV1 to PVp, the neural total prediction network DANN0 also obtains the reconstructed feature vector signals ROS1 to ROSp of all p obstacle objects HOP as input vectors. Therefore, the neural total prediction network DANN0 possesses information about both the prediction and reality.

[0578] The neural network DANN0 outputs the first virtually identified obstacle object VEO1, the second virtually identified obstacle object VEO2, and so on, up to the p-th virtually identified obstacle object VEOp. In this case, vector signals are involved, which are time- and / or spatially multiplexed to describe the identified obstacle objects. The values ​​of these signals can, for example, represent object type, object location, object orientation, and other object characteristics.

[0579] Initially, the obstacle object is unknown.

[0580] The first, second, and p-th prediction periods PZ1, PZ2, ..., PZp are briefly described below. The third through (p-1)-th prediction periods proceed similarly; however, this is not described in detail for clear reasons. Figure 27 As shown in the image.

[0581] First forecast period :

[0582] In the first prediction period PZ1, the neural network DANN0 identifies a first virtual obstacle object VEO1. The first virtual obstacle object VEO1 is stored in the first buffer memory VM1 of the model portion. The first artificial neural single prediction network MNN1 of the model portion reconstructs the reconstructed first feature vector signal RV11 of the first virtual obstacle object from the first virtual obstacle object VEO1. The reconstructed first feature vector signal RV11 of the first virtual obstacle object is stored in the first prediction memory SPRV1. In the first prediction period of the next prediction period, the currently calculated reconstructed first feature vector signal RV11 of the first virtual obstacle object is subtracted from the stored reconstructed first feature vector signal RV11 of the first virtual obstacle object at the summer. The result is the reconstructed second feature vector signal RV12 of the first virtual obstacle object. Therefore, this result represents the deviation from the previous prediction period. Then, the current reconstructed first feature vector signal RV11 of the first virtual obstacle object overwrites the contents of the first prediction memory SPRV1. In this way, the time-varying characteristics required for computational speed are considered in the prediction. Next, as described, the reconstructed second feature vector signal RV12 of the first virtual obstacle object is subtracted from the reconstructed first feature vector signal ROS1 of the first obstacle object and stored in the obstacle object memory SP1, thereby obtaining the first predicted feature vector signal PV1. Thus, the first prediction cycle ends.

[0583] Second forecast period :

[0584] In the second prediction period PZ2, the neural network DANN0 identifies the second virtual obstacle object VEO2. The second virtual obstacle object VEO2 is stored in the second buffer memory VM2 of the model part. The second artificial neural single prediction network MNN2 of the model part reconstructs the reconstructed first feature vector signal RV21 of the second virtual obstacle object from the second virtual obstacle object VEO2. The reconstructed second feature vector signal RV21 of the first virtual obstacle object is stored in the second prediction memory SPRV2. In the second prediction period of the next prediction period, the currently calculated reconstructed first feature vector signal RV21 of the second virtual obstacle object is subtracted from the stored reconstructed first feature vector signal RV21 of the second virtual obstacle object at the summer. The result is the reconstructed second feature vector signal RV22 of the second virtual obstacle object. Therefore, this result represents the deviation from the previous prediction period. Then, the current reconstructed first feature vector signal RV21 of the second virtual obstacle object overwrites the contents of the second prediction memory SPRV2. In this way, the time-varying characteristics required for computational speed are considered in the prediction. Next, as described, the reconstructed second feature vector signal RV22 of the second virtual obstacle object is subtracted from the reconstructed first feature vector signal ROS2 of the second obstacle object stored in the obstacle object memory SP1, thereby obtaining the second predicted feature vector signal PV2. This concludes the second prediction cycle.

[0585] The nth prediction period :

[0586] In the p-th prediction period PZp, the overall prediction network DANN0 identifies the p-th virtually identified obstacle object VEOo. The p-th virtually identified obstacle object VEOo is stored in the p-th buffer memory VMp of the model part. The p-th artificial neural single prediction network MNNp of the model part reconstructs the reconstructed first feature vector signal RVp1 of the p-th virtual obstacle object from the p-th virtually identified obstacle object VEOo. The reconstructed first feature vector signal RVp1 of the p-th virtual obstacle object is stored in the p-th prediction memory SPRVp. In the p-th prediction period of the next prediction period, the currently calculated reconstructed first feature vector signal RVp1 of the p-th virtual obstacle object is subtracted from the stored reconstructed first feature vector signal RVp1 of the p-th virtual obstacle object at the summer. The result is the reconstructed second feature vector signal RVp2 of the p-th virtual obstacle object. Therefore, this result represents the deviation from the previous prediction period. Next, the reconstructed first feature vector signal RVp1 of the p-th virtual obstacle object overwrites the contents of the p-th prediction memory SPRVp. This takes into account, for example, the time-varying characteristics required for computational speed, in the prediction. Then, as described, the reconstructed second feature vector signal RVp2 of the p-th virtual obstacle object is subtracted from the reconstructed first feature vector signal ROSp of the p-th obstacle object stored in the obstacle object memory SP1, thereby obtaining the p-th prediction feature vector signal PVp. Thus, the p-th prediction cycle ends.

[0587] The artificial neural reality simulation network MNN0 in the model part reconstructs the virtual reality simulation feature vector signal VRV based on the first virtual recognized obstacle object VEO1, the second virtual recognized obstacle object VEO2, and all other virtual recognized obstacle objects VEO3 to VEOp-1 up to the p-th virtual recognized obstacle object VEOp (including the p-th virtual recognized obstacle object VEOp). The reality simulation feature vector signal can be, for example, a multi-dimensional environment map.

[0588] The simulated characteristic vector signal VRV and the associated memories VM1, VM2, ..., VMp can be read at any time. These memories contain evaluable information about the spatial location of the object identified at the time points between emitted ultrasonic pulses.

[0589] Subsequently, according to Figure 28 This study explores the time synchronization of each step in predicting changes in detected obstacle objects. Figure 28The upper portion of the diagram shows the successive measurement or identification time periods ΔT on the t-axis. When this invention is applied to an ultrasonic measurement system, an ultrasonic transmission signal is generated for each time period ΔT, followed by oscillation attenuation of the ultrasonic transducer, which performs both transmission and reception functions. This oscillation attenuation phase is then followed by a reception phase for each time period ΔT.

[0590] The prediction time period can be roughly understood as having the same duration as the aforementioned time period ΔT. The prediction time period EP is divided into individual prediction periods PP1, PP2, ..., PPq, where q is a natural number greater than 2. Within each prediction period, there are prediction cycles PZ1, PZ2, ..., PZp, which are equal to the number of identified obstacle objects. Predictions of changes in the identified obstacle objects can begin as early as the second time period ΔT. Since the prediction is based on "experience from the past," it is appropriate to begin the prediction at a slightly later point in time, but not too late. Ultimately, this depends on the application.

[0591] Alternatively, the invention can also be rewritten using the feature groups mentioned below, wherein these feature groups can be arbitrarily combined with each other and each feature of a feature group can be combined with one or more features of one or more other feature groups and / or one or more of the previously described design schemes.

[0592] 1. A method for transmitting sensor data, particularly sensor data from an ultrasonic sensor, from a sensor to a computer system, particularly a computer system in a vehicle, the method comprising or including the following steps:

[0593] - Emits a burst of ultrasonic pulses;

[0594] - Receives ultrasonic signals and generates ultrasonic receiving signals;

[0595] - The characteristic vector signal 138 is formed based on the ultrasonic received signal 1 or the modified residual signal 660 derived therefrom;

[0596] - Evaluate at least 138 time segments of the eigenvector signal (Hamming-Windows).

[0597] - This is achieved by forming at least one binary, digital, or analog distance value between the feature vector signal and one or more prototype values ​​of the signal object for an identifiable signal object category, and assigning the identifiable signal object category as the identified signal object if the value of this distance value is numerically lower than one or more predetermined binary, digital, or analog distance values, and / or

[0598] - The method is particularly by means of estimator 151, using a neural network model and / or Petri network to assign identifiable signal object categories as identified signal objects;

[0599] - Within the ultrasound received signal 1, identify signal objects and classify these signal objects into the assigned signal object category.

[0600] - One of the signal object categories can also include only one signal object, and

[0601] - Wherein, each signal object 122 identified and classified is assigned at least one assigned signal object parameter and a symbol corresponding to the signal object category assigned to that signal object, or

[0602] - Wherein, for each thus identified and classified signal object 122, at least one assigned signal object parameter and a symbol for that signal object are determined.

[0603] - Transmit at least the symbol of the identified signal object category 122 and at least the assigned signal object parameter of the identified signal object category 122.

[0604] 2. According to the method described in item 1,

[0605] In a point-to-point connection, the symbols of signal object categories identified within the sensor and at least one assigned signal object parameter of each of these signal object categories are transmitted from the sensor to the computer system in a temporally prioritized manner, in terms of the temporal transmission order of the symbols of the identified signal object categories and the signal object parameters of each of these identified signal object categories.

[0606] 3. According to the method described in item 1 or 2,

[0607] The method includes the following additional steps: determining a chirp value as an assigned signal object parameter, the chirp value indicating whether the identified signal object is an echo of an ultrasonic transmission pulse train with chirped up, chirped down, or no chirp characteristics.

[0608] 4. The method described in one or more of items 1 to 3,

[0609] This involves using neural network models for classification.

[0610] 5. The method described in one or more of items 1 to 4,

[0611] This involves using neural network models for classification.

[0612] 6. The method described in one or more of items 1 to 5,

[0613] The classification is performed using an Hidden Markov Model (HMM).

[0614] 7. The method described in one or more of items 1 to 5,

[0615] The method includes the additional step of generating a confidence signal by forming a correlation between a signal received or derived on one side and a reference signal on the other side.

[0616] 8. The method described in one or more of items 1 to 7,

[0617] The method includes the following additional step: generating a phase signal.

[0618] 9. According to the method described in item 8,

[0619] The method includes the additional step of generating a phase position signal by forming a correlation between a phase signal or a signal derived therefrom and a reference signal.

[0620] 10. According to the method described in item 9,

[0621] The method includes the additional step of comparing the phase position signal with one or more thresholds to generate a discretized phase position signal.

[0622] 11. The method described in one or more of items 1 to 10,

[0623] At least one of the signal object categories is wavelet.

[0624] 12. According to the method described in item 11,

[0625] At least one of the wavelets is a triangular wavelet.

[0626] 13. According to the method described in item 11 or 12,

[0627] At least one of the wavelets is a rectangular wavelet.

[0628] 14. The method described in one or more of items 11 to 13,

[0629] At least one of the wavelets is a sinusoidal half-wavelet.

[0630] 15. The method described in one or more of items 11 to 14,

[0631] One of the parameters of the signal object is:

[0632] - The time offset of the wavelet of the identified signal object, or

[0633] - Time compression or expansion of the wavelet of the identified signal object, or

[0634] - The amplitude of the wavelet of the identified signal object.

[0635] 16. The method described in one or more of items 1 to 15,

[0636] The transmission of sensor fault status

[0637] - Relative to the transmission of at least one identified signal object category and / or

[0638] - In relation to the transmission of parameters of the assigned signal object.

[0639] They should be carried out in a higher priority order.

[0640] 17. The method described in one or more of items 1 to 16,

[0641] The signal object includes a combination of two, three, four or more basic signal objects.

[0642] 18. According to the method described in item 17,

[0643] The basic signal object is the intersection of the amplitude of the envelope signal 1 and the threshold signal SW at the intersection time point.

[0644] 19. The method described in accordance with paragraph 17 or 18,

[0645] The basic signal object is the intersection of the amplitude of the envelope signal 1 and the threshold signal SW in the rising direction at the intersection time point.

[0646] 20. The method described in one or more of items 17 to 19,

[0647] The basic signal object is the intersection of the amplitude of the envelope signal 1 and the threshold signal SW in the descent direction at the intersection time point.

[0648] 21. The method described in one or more of items 17 to 20,

[0649] The basic signal object is the maximum value of the amplitude of the envelope signal 1 at the maximum time point, which is higher than the value of the threshold signal SW.

[0650] 22. The method described in one or more of items 17 to 21,

[0651] The basic signal object is the minimum value of the amplitude of the envelope signal 1 at the minimum time point that is higher than the value of the threshold signal SW.

[0652] 23. The method described in one or more of items 17 to 22,

[0653] The basic signal object is a predefined time series and / or time grouping of other basic signal objects.

[0654] 24. The method described in one or more of items 1 to 23,

[0655] The transmission of at least the symbol of the identified signal object category and the at least one assigned signal object parameter of the identified signal object category is the transmission of the signal object category of a signal object that is a predefined time series of other signal objects, wherein at least one signal object category of at least one of these other signal objects is not transmitted.

[0656] 25. The method described in one or more of items 1 to 24,

[0657] This method has the following additional steps:

[0658] - Reconstruct the reconstructed ultrasonic echo signal model 610 based on the identified signal object 122.

[0659] - Subtract the reconstructed ultrasound echo signal model 610 from the ultrasound received signal 1 to form the residual signal 660.

[0660] - Use the residual signal 660 to form the eigenvector signal 138.

[0661] 26. The method described in one or more of items 1 to 25,

[0662] The method has the following additional step: if the value of the sampled value of the residual signal 660 is lower than the value of a pre-given threshold curve, the classification on the signal object is terminated.

[0663] 27. A decompression method, particularly for decompressing ultrasonic received data compressed by means of one or more of the methods according to claims 1 to 26, the decompression method comprising the following steps:

[0664] - Receive the data to be decompressed.

[0665] - Provides no signal ( Figure 13a) Ultrasonic echo signal model,

[0666] - By adding the data included in the ultrasound reception and meeting pre-given conditions ( Figure 13 (b) Signal object 160 is used to supplement the ultrasonic echo signal model, where it is possible without pre-given conditions. Figure 15 and 16 ),

[0667] - A reconstructed ultrasound received signal is generated, which can be achieved in the form of sampled values ​​forming a reconstructed envelope.

[0668] - Use the reconstructed ultrasound receiving signal.

[0669] 28. The decompression method according to item 27,

[0670] The decompression method has the following steps: using the reconstructed ultrasonic received signal in the vehicle for autonomous driving purposes and / or for environmental map generation purposes.

[0671] 29. A sensor, particularly an ultrasonic sensor, adapted for or configured to perform the method according to one or more of claims 1 to 26.

[0672] 30. A computer system,

[0673] - The computer system is suitable for and configured to perform the methods described in accordance with paragraph 27 or 28, or

[0674] - It is configured for use with the sensor described in Item 29 and is suitable for obtaining data from the sensor.

[0675] 31. A sensor system having:

[0676] - At least one computer system as described in item 30; and

[0677] - At least one sensor as described in item 29,

[0678] - Wherein the computer system is suitable for and / or configured to perform the method described in accordance with paragraph 27 or 28.

[0679] 32. A sensor system having:

[0680] - At least one computer system as described in item 30; and

[0681] - At least two sensors as described in item 29,

[0682] - Wherein the sensor is configured such that data transmission between these sensors and the computer system is performed or is capable of being performed according to the method described in items 1 to 28.

[0683] 33. The sensor system according to item 32,

[0684] Within these sensors, each ultrasonic receiving signal, i.e., at least two ultrasonic receiving signals, is compressed and transmitted to the computer system by means of a method corresponding to one or more of the methods described according to items 1 to 26, and

[0685] - Within the computer system, the at least two ultrasonic received signals are reconstructed into reconstructed ultrasonic received signals by decompressing the data received from the sensors.

[0686] 34. The sensor system according to item 32,

[0687] The computer system uses reconstructed ultrasonic received signals to perform object recognition of objects in the environment of these sensors.

[0688] 35. The sensor system according to item 34,

[0689] The computer system uses reconstructed ultrasonic received signals and additional signals from other sensors, particularly radar sensors, to perform object recognition of objects in the environment of these sensors.

[0690] 36. The sensor system according to claim 34 or 35,

[0691] The computer system creates an environmental map of the sensors or devices that are part of the sensors based on the identified objects.

[0692] 37. An ultrasonic sensor system ( Figure 7 ), which has:

[0693] - Feature extractor 111;

[0694] - Estimators 150 and 151;

[0695] - Physical interface 101; and

[0696] - Ultrasonic transducer 100,

[0697] - Wherein the ultrasonic transducer 100 is configured and / or set to receive ultrasonic signals and thereby form an ultrasonic transducer signal 102.

[0698] - Wherein the feature vector extractor 111 is configured and / or set to form a feature vector signal 138 based on the ultrasound transducer signal 102 or a signal derived therefrom, wherein the ultrasound sensor system is configured and set to: identify signal objects in the ultrasound received signal and / or the ultrasound echo signal 1 derived therefrom by means of estimators 151, 150 and classify these signal objects into signal object categories.

[0699] - One of the signal object categories can also include only one signal object, and

[0700] - Wherein, each signal object 122 identified and classified is assigned at least one assigned signal object parameter and a symbol corresponding to the signal object category assigned to that signal object, or

[0701] - Wherein, for each thus identified and classified signal object 122, at least one assigned signal object parameter and a symbol for that signal object are determined, and

[0702] - The at least one symbol of the identified signal object category 122 and the at least one assigned signal object parameter of the identified signal object category 122 are transmitted to the upper-level computer system via the data bus.

[0703] 38. The ultrasonic sensor system according to claim 37,

[0704] The estimator 150 includes a distance determiner 112 and a prototype database 115.

[0705] 39. The ultrasonic sensor system according to claim 38,

[0706] The estimator 150 includes a Viterbi estimator and a signal object database.

[0707] 40. An ultrasonic sensor system according to one or more of points 37 to 39,

[0708] Estimator 150 uses a neural network model.

[0709] 41. A method for operating an ultrasonic sensor, the method comprising the following steps:

[0710] - Waiting for an ultrasonic receiving signal that is different from the background noise;

[0711] - If the received ultrasound signal is different from the background noise, the system transitions to the state "no signal object prototype" and executes methods for identifying and classifying the signal prototype.

[0712] - If the first signal basic prototype of a pre-given signal basic prototype sequence is identified, then the sequence of states of the signal basic prototype assigned to that sequence is transformed.

[0713] - Follow the basic sequence of signal prototypes until the basic sequence of signal prototypes ends;

[0714] - If the signal prototype sequence ends, the existence of a signal prototype in the signal prototype sequence is inferred and the signal object assigned to the sequence is inferred, and the signal object is notified by a signal;

[0715] - Disrupt the sequence if a timeout occurs and / or if the basic prototype of the signal is detected once or multiple times at a location in the sequence where the basic prototype of the signal is not expected to be present, or at a next location where the basic prototype of the signal is not expected to be present.

[0716] - Return to the state "No Signal Object Basic Prototype".

[0717] 42. A method for operating an ultrasonic sensor, the method comprising the following steps:

[0718] - Detect the received ultrasound signal;

[0719] - Provides a signal-free ultrasound echo signal model 610 ( Figure 15 a)

[0720] - Perform the following steps at least once, or optionally repeatedly:

[0721] - Subtract the reconstructed ultrasound echo signal model 610 from the ultrasound received signal 1 to form the residual signal 660.

[0722] - Execute a method for identifying signal objects in residual signal 660.

[0723] - Give the ultrasonic echo signal model 610 ( Figure 15 (b to 16h) Supplement the signal waveforms of the identified signal objects 600 to 605;

[0724] - If the absolute value of the residual signal (Betragwert) is lower than the absolute value of a pre-given threshold signal, then the repetition of these steps ends;

[0725] - Transmit symbols for at least a portion of the identified signal objects and optionally use this information.

[0726] 43. An ultrasonic sensor system, particularly an ultrasonic sensor system for use in a vehicle, the ultrasonic sensor system having:

[0727] - First ultrasonic sensor;

[0728] - At least one second ultrasonic sensor; and

[0729] - Computer system,

[0730] - Each of the at least two ultrasonic sensors implements a method for transmitting sensor data from the respective ultrasonic sensor to a computer system, the method comprising the following steps:

[0731] - Emits an ultrasonic pulse train α, and

[0732] - Receives ultrasonic signals and forms an ultrasonic received signal β, and

[0733] - Data compression of the ultrasonic received signal is performed using the reconstructed ultrasonic echo signal model 610 to generate compressed data γ, and

[0734] - Transfer these compressed data to the computer system δ, and

[0735] - Within this computer system, at least two compressed ultrasonic received signals are processed using a reconstructed ultrasonic echo signal model ( Figure 15 and 16 The ultrasound received signal is decompressed or reconstructed, and

[0736] - The computer system uses these reconstructed ultrasonic received signals to perform object recognition of objects in the environment of the ultrasonic sensor.

[0737] 44. The ultrasonic sensor system according to claim 43, wherein the computer system performs object recognition of objects in the sensor's environment by means of reconstructed ultrasonic received signals and additional signals from other sensors, particularly radar sensors.

[0738] 45. The ultrasonic sensor system according to claim 43 or 44, wherein the computer system creates an environmental map of the sensor or a device in part being the sensor based on the identified objects.

[0739] 46. ​​A method for transmitting sensor data, particularly sensor data from an ultrasonic sensor, from a sensor to a computer system, particularly a computer system in a vehicle, the method comprising or including the following steps:

[0740] - Emits an ultrasonic pulse train α;

[0741] - Receives ultrasonic signals and generates received signals, especially ultrasonic received signals β;

[0742] - Perform data compression on the received signal to produce compressed data γ.

[0743] - Using neural network model 151 ( Figure 12 ) and / or

[0744] - Using HMM model ( Figure 9 and Figure 7 ) and / or

[0745] - By means of Petri network ( Figure 12 ) and / or

[0746] - Using the reconstructed ultrasound echo signal model 610,

[0747] - Transmit these compressed data to the computer system.

[0748] 47. According to the method described in item 46,

[0749] Data transmission is achieved via a bidirectional single-wire data bus.

[0750] - The sensor transmits data to the computer system in a current-modulated manner, and

[0751] The computer system sends data to the sensor in a voltage-modulated manner.

[0752] 48. The method described in paragraph 46 or 47,

[0753] The feature is that the PSI5 data bus and / or DSI3 data bus are used for data transmission.

[0754] 49. The method described in one or more of items 46 to 48,

[0755] The data is transmitted to the computer system at a transmission rate of >200 kBit / s and from the computer system to at least one sensor at a transmission rate of >10 kBit / s, preferably 20 kBit / s.

[0756] 50. The method described in one or more of items 46 to 49,

[0757] In order to transmit data from the sensor to the computer system, the transmitting current is modulated onto the data bus, and the current intensity of the transmitting current is < 50 mA, preferably < 5 mA.

[0758] 51. A sensor, particularly an ultrasonic sensor, adapted for or configured to perform the method described in one or more of claims 46 to 50.

[0759] 52. A computer system adapted or configured to perform the method described in one or more of claims 46 to 50.

[0760] 53. A sensor system having:

[0761] - At least one computer system as described in claim 52; and

[0762] - At least two sensors as described in item 51,

[0763] - Wherein the sensor is configured such that data transmission between these sensors and the computer system is performed or is capable of being performed according to one or more of the methods described in items 46 to 50.

[0764] 54. The sensor system according to item 53,

[0765] - Within these sensors, each ultrasonic receiving signal, i.e., at least two ultrasonic receiving signals, is compressed and transmitted to the computer system by means of a method corresponding to one or more of the methods described according to items 46 to 50, and

[0766] - Within the computer system, the at least two ultrasonic received signals are reconstructed into reconstructed ultrasonic received signals.

[0767] 55. The sensor system according to item 54,

[0768] The computer system uses reconstructed ultrasonic received signals to perform object recognition of objects in the environment of these sensors.

[0769] 56. The sensor system according to item 55,

[0770] The computer system uses reconstructed ultrasonic received signals and additional signals from other sensors, particularly radar sensors, to perform object recognition of objects in the environment of these sensors.

[0771] 57. The sensor system according to claim 55 or 56,

[0772] The computer system creates an environmental map of the sensors or devices that are part of the sensors based on the identified objects.

[0773] List of reference numerals

[0774] (against Figures 1 to 20 )

[0775] α emits an ultrasonic pulse train.

[0776] β receives the ultrasonic pulse train reflected at the obstacle object and converts it into an electrical received signal.

[0777] γ compresses the received electrical signal

[0778] γa samples the electrical input signal and forms a sampled electrical input signal, wherein preferably, a timestamp is assigned to each sampled value of the electrical input signal.

[0779] γb, for example, uses a matched filter to determine multiple spectral values ​​for the prototype signal object category. These multiple spectral values ​​collectively form a feature vector. This formation is preferably performed continuously, resulting in a stream of feature vector values. Preferably, each feature vector value can be assigned a timestamp again.

[0780] γc optionally, but preferably, performs normalization of the spectral coefficients of the corresponding eigenvectors of the timestamp values ​​before associating them with the prototype signal object category in the form of pre-given prototype eigenvector values ​​from the prototype library.

[0781] γd determines the distance between the current feature vector value and the value of the prototype signal object category in the form of pre-given prototype feature vector values ​​from the prototype library.

[0782] γe selects the most similar prototype signal object category, in the form of pre-given prototype feature vector values ​​from the prototype library, having a preferred minimum distance from the current feature vector, and uses the symbol of this signal object category, along with its timestamp value, as the compressed data. If necessary, other data, particularly signal object parameters and their amplitudes, can also be used as compressed data. This compressed data then forms the compressed received signal.

[0783] δ transmits the compressed electrical received signal to the computer system.

[0784] 1. The received ultrasonic signal, also known here as the envelope of the ultrasonic echo signal.

[0785] 2. The output signals (transmitted information) of the existing IO interface.

[0786] 3. Information transmitted according to the existing LIN interface

[0787] 4. The first intersection point of the ultrasound echo signal 1 and the threshold signal SW in the downward direction

[0788] 5. The first intersection point of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction.

[0789] 6. The first maximum value of the ultrasonic echo signal 1 above the threshold signal SW

[0790] 7. The second intersection point of the ultrasound echo signal 1 and the threshold signal SW in the downward direction.

[0791] 8. The second intersection point of the ultrasound echo signal 1 and the threshold signal SW in the upward direction.

[0792] 9. The second maximum value of the ultrasound echo signal 1 above the threshold signal SW

[0793] The first minimum value of the ultrasonic echo signal 1 above the threshold signal SW

[0794] 11. The third maximum value of the ultrasound echo signal 1 above the threshold signal SW

[0795] 12. The third intersection point in the downward direction of the ultrasonic echo signal 1 and the threshold signal SW

[0796] 13. The third intersection point in the upward direction of the ultrasonic echo signal 1 and the threshold signal SW.

[0797] 14. The fourth maximum value of the ultrasound echo signal 1 above the threshold signal SW

[0798] 15. The fourth intersection point in the downward direction of the ultrasound echo signal 1 and the threshold signal SW.

[0799] The fourth intersection point of the 16 ultrasound echo signal 1 and the threshold signal SW in the upward direction.

[0800] 17. The fifth maximum value of the ultrasound echo signal 1 above the threshold signal SW

[0801] 18. The fifth intersection point of the ultrasound echo signal 1 and the threshold signal SW in the downward direction.

[0802] 19. The fifth intersection point of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction.

[0803] The sixth maximum value of the ultrasound echo signal 1 above the threshold signal SW.

[0804] 21. The sixth intersection point in the downward direction of the ultrasound echo signal 1 and the threshold signal SW.

[0805] 22. The sixth intersection point of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction.

[0806] 23. The seventh maximum value of the ultrasound echo signal 1 above the threshold signal SW

[0807] 24. The seventh intersection point in the downward direction of the ultrasound echo signal 1 and the threshold signal SW.

[0808] 25. Envelope during an ultrasound pulse train

[0809] 26. Data of the first intersection point 4 in the downward direction of the ultrasonic echo signal 1 and the threshold signal SW is transmitted via a preferably bidirectional data bus.

[0810] 27. Data of the first intersection point 5 of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction is transmitted via a preferably bidirectional data bus.

[0811] 28. Data of the first maximum value 6 of the ultrasonic echo signal 1 above the threshold signal SW is transmitted via a preferably bidirectional data bus.

[0812] 29. Data of the second intersection point 7 of the ultrasonic echo signal 1 and the threshold signal SW in the downward direction is transmitted via a preferably bidirectional data bus.

[0813] 30 transmits data of the second intersection point 8 of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction via a preferably bidirectional data bus.

[0814] 31. Data of the second maximum value 9 of the ultrasonic echo signal 1 above the threshold signal SW and data of the first minimum value 10 of the ultrasonic echo signal 1 above the threshold signal SW are transmitted via a preferably bidirectional data bus.

[0815] 32. Data of the third maximum value 11 of the ultrasonic echo signal 1 above the threshold signal SW is transmitted via a preferred bidirectional data bus.

[0816] 34. Data of the third intersection point 12 of the ultrasonic echo signal 1 and the threshold signal SW in the downward direction is transmitted via a preferably bidirectional data bus.

[0817] 35. Data of the third intersection point 13 of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction is transmitted via a preferred bidirectional data bus.

[0818] 36. Data of the fourth maximum value 14 of the ultrasonic echo signal 1 above the threshold signal SW is transmitted via a preferred bidirectional data bus.

[0819] 37. Data of the fourth intersection point 15 of the ultrasonic echo signal 1 and the threshold signal SW in the downward direction is transmitted via a preferably bidirectional data bus.

[0820] 38. Data of the fourth intersection point 16 of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction is transmitted via a preferred bidirectional data bus.

[0821] 39. Data of the fifth maximum value 17 of the ultrasonic echo signal 1 above the threshold signal SW is transmitted via a preferred bidirectional data bus.

[0822] 40 transmits data at the fifth intersection point 18 in the downward direction of the ultrasonic echo signal 1 and the threshold signal SW via a preferably bidirectional data bus.

[0823] 41. Data of the fifth intersection point 19 of the ultrasonic echo signal 1 and the threshold signal SW in the upward direction is transmitted via a preferably bidirectional data bus.

[0824] 42. Data of the sixth intersection point 21 of the ultrasonic echo signal 1 and the threshold signal SW in the downward direction is transmitted via a preferably bidirectional data bus.

[0825] 43. After reception is completed, the received echo data is transmitted on the LIN bus according to existing technology.

[0826] 44. Data is transmitted on a LIN bus according to existing technology before the ultrasonic pulse train is emitted.

[0827] 45. Data is transmitted via an I / O interface according to existing technology before the ultrasonic pulse train is emitted.

[0828] 46. ​​The impact of ultrasonic pulse trains on the output signal of I / O interfaces according to existing technology

[0829] 47. Signals 5, 6, and 7 of the first echo on the I / O interface according to the prior art.

[0830] 48. Signals 8, 9, 10, 11, and 12 of the second echo on the I / O interface according to the prior art.

[0831] 49. Signals of the third and fourth echoes 13, 14, and 15 on the I / O interface according to the prior art.

[0832] 50 The fifth echo signals 16, 17, and 18 on the I / O interface according to the prior art

[0833] 51 The sixth echo signals 19, 20, and 21 on the I / O interface according to the prior art

[0834] 52 The seventh echo signals 22, 23, and 24 on the I / O interface according to the prior art

[0835] 53. Start command from computer system to sensor via data bus

[0836] 54 Preferably, periodic automatic data transmission between the sensor and the computer system is based on the DSI3 standard.

[0837] 55 Diagnostic position after measurement cycle

[0838] 56. End of the ultrasonic pulse train (end of pulse train transmission). Preferably, the end of the ultrasonic pulse train coincides with point 4.

[0839] 57. Beginning of the ultrasonic pulse train (start of pulse train transmission)

[0840] 58. End of data transmission

[0841] 100 ultrasound transducers, which may, for example, include a separate ultrasound transmitter and a separate ultrasound receiver.

[0842] 101 is a physical interface used to drive the ultrasound converter 100 and to process the ultrasound converter signal 102 received from the ultrasound converter 100 into an ultrasound echo signal 1 for subsequent signal object classification.

[0843] 102 Ultrasonic Converter Signal

[0844] 111 Feature Vector Extractor 111

[0845] 112 Distance Determiner (or Classifier)

[0846] 113 Viterbi Estimator

[0847] 115 Prototype Database

[0848] 116 Signal Object Database

[0849] The signal object with signal object parameters identified by 122

[0850] 123 intermediate parameter signal

[0851] 125 Significance Enhancer

[0852] 126LDA matrix

[0853] 138 Feature-Vektor-Signal or Feature Vector Signal

[0854] The first prototype signal of the prototype database 115 has the following barycentric coordinates.

[0855] The second prototype signal of the prototype database 115 has the following barycentric coordinates.

[0856] The barycenter coordinates of the basic object of the third prototype signal in prototype database 115 (143).

[0857] The barycenter coordinates of the fourth prototype signal basic object in prototype database 115 (144 prototype database)

[0858] The current eigenvector signal value of eigenvector signal 138 within the overlap range of the scattering range of the two signal basic objects in prototype database 115 and the centroid coordinates 142 and 143.

[0859] The eigenvector signal value 146 is too far from the centroid coordinates 141, 142, 143, and 144 of the centroid of any signal primitive object prototype in the prototype database 115.

[0860] The unique signal basic object prototype 147 has a scattering range (threshold ellipsoid) around its centroid 141.

[0861] 148 The current eigenvector signal value lies within the scattering range (threshold ellipsoid) 147 around the centroid 141 of the unique signal primitive object prototype 141, and the current eigenvector signal value can be reliably identified by the distance determiner 112 and transferred to the Viterbi estimator 113 as the identified signal primitive object 121.

[0862] 150 estimators with HMM models

[0863] 151 Estimators with Neural Network Models

[0864] 160 is a first signal object with a triangular shape having a chirped downward A and a data transmission priority of 1.

[0865] 161 is a second signal object with a triangular shape having a chirp-up B and a data transmission priority of 2.

[0866] 162 is a third signal object (curbstone profile) with a bimodal shape having chirp-up B and data transmission priority 3.

[0867] 163 is a fourth signal object with a triangular shape having a chirped downward A and a data transmission priority of 4.

[0868] 164 is a fifth signal object with a triangular shape having a chirped downward A and a data transmission priority of 5.

[0869] 165 is a sixth signal object with a triangular shape having a chirp-up B and a data transmission priority of 6.

[0870] 166. Echo signal transmitted and decompressed by means of a signal object

[0871] 600 Reconstructor

[0872] 601 is a memory for the ultrasonic echo signal 1. This memory preferably includes data of the echoes of ultrasonic pulses or ultrasonic pulse trains.

[0873] Subtractor 602 is used to subtract the reconstructed sampled values ​​of the ultrasonic echo signal, calculated by reconstructor 600 to form the reconstructed ultrasonic echo signal model 610, from the sampled values ​​of the ultrasonic echo signal 1 stored in memory 601, in order to form a residual signal 660, which replaces the ultrasonic echo signal 1 as the alternative input signal of feature vector extractor 111.

[0874] 603 is a reconstruction memory for the reconstructed ultrasonic echo signal model 610. The reconstruction memory is typically implemented as part of the reconstructor 600.

[0875] 610 Reconstructed Ultrasonic Echo Signal Model

[0876] 660 Residual Signal. This residual signal represents the compression error. The more objects identified, the smaller the compression error becomes. Typically, compression is interrupted if all samples of the residual signal 660 are below the compression threshold curve. The corresponding interruption signal is not shown in the attached figure.

[0877] a is used to transmit information of received ultrasonic echoes by means of an I / O interface according to the prior art.

[0878] The signal object identified by A with chirping downwards

[0879] au "arbritrary units" = freely chosen units

[0880] The signal object identified by B with chirping upwards.

[0881] b is used to transmit information from the received ultrasonic echo via a LIN interface based on existing technology.

[0882] c is used to transmit information of the received ultrasonic echo, having an envelope (ultrasonic echo signal 1) for comparison, by means of the proposed method and the proposed apparatus.

[0883] d is used to transmit envelope-free information from received ultrasonic echoes by means of the proposed method and the proposed apparatus.

[0884] Schematic signal shape of e when transmitting received echo information using an I / O interface based on existing technology.

[0885] The amplitude of the envelope of the ultrasound signal received by En (ultrasound echo signal 1)

[0886] f is an illustrative signal shape when transmitting received echo information using a LIN interface based on existing technology.

[0887] The schematic signal shape of g when transmitting received echo information using a bidirectional data interface.

[0888] The memory address of SA is in memory 601 or rebuild memory 603. This memory address is typically related to the time since the ultrasonic pulse and / or ultrasonic pulse train was emitted.

[0889] SB sends pulse train

[0890] SW threshold

[0891] t time

[0892] T E Reception time. This reception time typically begins 56 seconds after the end of the emitted ultrasound pulse train. It's possible that reception may have started earlier. However, this could lead to issues requiring additional measures if necessary.

[0893] List of reference numerals

[0894] (against Figures 21 to 25 )

[0895] A0 is the zeroth summer.

[0896] A1 First Summer

[0897] A2 Second Summer

[0898] An nth summer

[0899] ANN0 Neural Total Disorder Recognition Network

[0900] DANN0 Neural Total Prediction Network

[0901] DB Data Bus

[0902] DR drive

[0903] The first signal object transmitted by E1

[0904] The second signal object transmitted by E2

[0905] EC1 corrected first eigenvector signal

[0906] EC2 modified second eigenvector signal

[0907] EC3 modified third eigenvector signal

[0908] ECm is the corrected m-th eigenvector signal.

[0909] ECU data processing unit

[0910] Reconstructed feature vector signal of the first sensor of EF1

[0911] Reconstructed feature vector signal of the second sensor of EF2

[0912] Reconstructed eigenvector signal of the EF3 third sensor

[0913] The reconstructed feature vector signal of the m-th sensor EFm

[0914] The first buffer memory of the EM1 data processing unit

[0915] The second buffer memory of the EM2 data processing unit

[0916] The nth buffer memory of the EMn data processing unit

[0917] The nth signal object transmitted by En

[0918] ENN1 First Neural Single Decompressed Network

[0919] ENN2 second neural single decompression network

[0920] ENNn, the nth neural single decompression network

[0921] The first buffer memory of the EO1 reality part

[0922] The second buffer memory of the EO2 reality section

[0923] The p-th buffer memory of the EOp reality part

[0924] EP identification period / predicted time period

[0925] EPS1 First Measurement Time Point

[0926] EPS2 Second Measurement Time Point

[0927] ER reconstructed eigenvector signal

[0928] The first eigenvector signal reconstructed from ER1

[0929] The reconstructed second eigenvector signal of ER2

[0930] The (n-1)th eigenvector signal reconstructed from ERn-1

[0931] The reconstructed nth eigenvector signal of ERn

[0932] The first summer of the ES1 data processing unit

[0933] The second summer of the ES2 data processing unit

[0934] The nth summer of the ESn data processing unit

[0935] The reconstructed first eigenvector signal of the EV1 reality portion

[0936] The reconstructed second eigenvector signal of the EV2 reality portion

[0937] The reconstructed third eigenvector signal of the EV3 reality portion

[0938] The reconstructed m-th eigenvector signal of the real part of EVm

[0939] F0 is the zeroth eigenvector signal.

[0940] F1 first eigenvector signal

[0941] F2 second eigenvector signal

[0942] FE Feature Extraction

[0943] IDB Internal Data Bus

[0944] IFE1 Inverse Feature Extraction

[0945] IFE2 Inverse Feature Extraction

[0946] IFEn-1 Inverse Feature Extraction

[0947] EFEn Inverse Feature Extraction

[0948] IM0 Zeroth Buffer

[0949] IM1 First Buffer Memory

[0950] IM2 Second Buffer Memory

[0951] IMn, the nth buffer memory

[0952] MNN0 Neural Reality Simulation Network

[0953] MNN1 First Neural Single Prediction Network

[0954] MNN2 Second Neural Single Prediction Network

[0955] MNNp, the p-th neural single prediction network

[0956] The first summer in the MS1 model section

[0957] The second summer in the MS2 model section

[0958] The p-th summer in the MSp model section

[0959] NN0 Neural Signal Object Recognition Network

[0960] NN1 First Neural Single Reconstruction Network

[0961] NN2 Second Neural Single Reconstruction Network

[0962] NNn nth neural single reconstruction network

[0963] The first signal object identified by O1

[0964] The second signal object identified by O2

[0965] The nth signal object identified by On

[0966] PP1 First Forecast Period

[0967] PP2 Second Forecast Period

[0968] PPm in the m-th prediction period

[0969] PV1 first predicted eigenvector signal

[0970] PV2 second predicted eigenvector signal

[0971] PVp, the p-th predicted feature vector signal

[0972] PZ1 First Prediction Period

[0973] PZ2 second prediction period

[0974] PZn prediction period n

[0975] R1 reconstructed first eigenvector signal

[0976] R2 reconstructed second eigenvector signal

[0977] RC identification control

[0978] REK Reconstructor

[0979] RF reconstructed eigenvector signal

[0980] The (n-1)th eigenvector signal reconstructed from Rn-1

[0981] The reconstructed nth eigenvector signal of Rn

[0982] The first neural network for recognizing single obstacles in the real-world part of RNN1.

[0983] The second neural network for recognizing single obstacles in the real-world part of RNN2.

[0984] The p-th neural network for recognizing single obstacles in the real-world part of RNNp.

[0985] RO1 is the reconstructed obstacle object feature vector signal for the first object across all m sensors.

[0986] RO2 is the reconstructed obstacle object feature vector signal for the second object from all m sensors.

[0987] RO11 reconstructed obstacle object feature vector signal for the first object of the first sensor.

[0988] RO12 for the reconstructed obstacle object feature vector signal of the first object from the second sensor

[0989] RO13 reconstructed obstacle object feature vector signal for the first object from the third sensor

[0990] RO1m is the reconstructed obstacle object feature vector signal for the first object of the m-th sensor.

[0991] RO21 reconstructed obstacle feature vector signal of the second object for the first sensor

[0992] RO22 reconstructs the obstacle feature vector signal of the second object for the second sensor.

[0993] RO23 reconstructs obstacle feature vector signals for the second object from the third sensor.

[0994] RO2m is the reconstructed obstacle feature vector signal of the second object for the m-th sensor.

[0995] ROp is the reconstructed obstacle feature vector signal for the p-th object from all m sensors.

[0996] ROp1 is the reconstructed obstacle feature vector signal for the p-th object of the first sensor.

[0997] ROp2 is the reconstructed obstacle feature vector signal for the p-th object of the second sensor.

[0998] ROp3 is the reconstructed obstacle feature vector signal for the p-th object of the third sensor.

[0999] ROpm is the reconstructed obstacle feature vector signal for the p-th object from the m-th sensor.

[1000] The reconstructed and stored feature vector signal of the first obstacle object in ROS1

[1001] The reconstructed and stored feature vector signal of the second obstacle object in ROS2.

[1002] The reconstructed and stored feature vector signal of the p-th obstacle object in ROSp.

[1003] The first summer of the RS1 real-world part

[1004] The second summer in the RS2 real-world part

[1005] RS3 Reality Third Summer

[1006] The m-th summer in the reality part of RSm

[1007] RSP Rebuild Memory

[1008] The reconstructed feature vector signal of the obstacle object first virtually identified by RV11

[1009] The reconstructed feature vector signal of the obstacle object first virtually identified by RV12

[1010] The reconstructed feature vector signal of the obstacle object recognized by RV21 second virtual recognition

[1011] The reconstructed feature vector signal of the obstacle object recognized by RV22 second virtual recognition

[1012] The reconstructed feature vector signal of the p-th virtually identified obstacle object in RVp1

[1013] The reconstructed feature vector signal of the p-th virtually identified obstacle object in RVp2

[1014] RX receiver

[1015] S1 First Sensor

[1016] S2 Second Sensor

[1017] S3 Third Sensor

[1018] Sm is the m-th sensor

[1019] SCEECU system control

[1020] SCU Ultrasonic System Control

[1021] SP1 First Obstacle Object Memory

[1022] SP2 Second Obstacle Object Memory

[1023] SPC memory control

[1024] SPp, the memory of the p-th obstacle object.

[1025] SPRV1 First Predictive Memory

[1026] SPRV2 Second Predictive Memory

[1027] SPRVp Prediction Memory p

[1028] TR Ultrasonic Converter

[1029] TEC transmitter controller

[1030] TRU data bus interface

[1031] TU ultrasonic transmitter control device

[1032] TRE data bus interface

[1033] VEO1 first virtual recognition of obstacle objects

[1034] VEO2 second virtual recognition obstacle object

[1035] The p-th virtually recognized obstacle object in VEOp

[1036] The first buffer memory of the VM1 model section

[1037] The second buffer memory of the VM2 model section

[1038] The p-th buffer memory in the VMp model part

[1039] VRV Reality Simulation Feature Vector Signal

[1040] Bibliography

[1041] DE 4433957 A1

[1042] DE 102012015967 A1

[1043] DE 102011085286 A1

[1044] DE 102015104934 A1

[1045] DE 102010041424 A1

[1046] DE 102013226373 A1

[1047] DE 10024959 A1

[1048] DE 102013015402 A1

[1049] DE 102018106244 B3

[1050] DE 102019106190 A1

[1051] DE 102017123049 B3

[1052] DE 102017123050 B3

[1053] DE 102017123051 B3

[1054] DE 102017123052 B3

[1055] WO 2012 / 016834 A1

[1056] WO 2014 / 108300 A1

[1057] WO 2018 / 210966 A1

[1058] US 2006 / 0250297 A1

Claims

1. A method for identifying the presence of an obstacle object in a detection area based on sensor signals provided by a plurality of sensors that detect the area, wherein in the method... A) For each sensor, signal waveform characteristics are extracted from the sensor signal using feature extraction, and a feature vector signal (EF1, EF2, EF3, ..., EFm) representing the sensor signal is formed based on the features. B) The feature vector signals (EF1, EF2, EF3, ..., EFm) are input to the artificial neural obstacle object recognition network (ANN0). The artificial neural obstacle object recognition network identifies at least one obstacle object based on the features of the feature vector signals (EF1, EF2, EF3, ..., EFm) and stores the description of each obstacle object in a separate obstacle object memory (EO1, EO2, ..., EOp). C) The information from each obstacle object memory (EO1, EO2, ..., EOp) is input as input data into the artificial neural single obstacle object recognition network (RNN1, RNN2, ..., RNNp) allocated to the obstacle object memory. D) Each neural obstacle object recognition network (RNN1, RNN2, ..., RNNp) outputs obstacle object feature vector signals (RO1, RO2, ..., ROp) representing the obstacle objects in the associated obstacle object memory (EO1, EO2, ..., EOp) and intermediate feature vector signals (ROij, where i = 1, 2, ..., p, where p equals the number of obstacle objects, and j = 1, 2, ..., m, where m equals the number of sensors), each of the intermediate feature vector signals being assigned a unique sensor (S1, S2, S3, ..., Sm). E) For each sensor (S1, S2, S3, ..., Sm), the intermediate feature vector signals (ROij, where i = 1, 2, ..., p, where p equals the number of obstacle objects, and j = 1, 2, ..., m, where m equals the number of sensors) output by the neural single obstacle object recognition network (RNN1, RNN2, ..., RNNp) are summed to form the corrected feature vector signal (EV1, EV2, EV3, ..., EVm). F) For each sensor (S1, S2, S3, ..., Sm), form a residual feature vector signal (EC1, EC2, EC3, ..., ECm) by subtracting the modified feature vector signal (EV1, EV2, EV3, ..., EVm) from the feature vector signal (EF1, EF2, EF3, ..., EFm). G) If the residual feature vector signal (EC1, EC2, EC3, ..., ECm) is greater than the threshold signal, repeat steps B) to F) using the corresponding updated residual feature vector signal (EC1, EC2, EC3, ..., ECm). H) Otherwise, the obstacle object feature vector signals (RO1, RO2, ..., RNNp) output by the neural sub-networks (RNN1, RNN2, ..., RNNp) represent the obstacle objects respectively, and the potential obstacle objects in the detection area are determined based on the obstacle object feature vector signals (RO1, RO2, ..., RNNp).

2. The method according to claim 1, characterized in that, The feature vector signals (EF1, EF2, EF3, ..., EFm) assigned to the relevant sensors (S1, S2, S3, ..., Sm) are provided as a reconstructed total feature vector signal (ER) formed according to a method for decompressing compressed sensor signal data describing the sensor signals of the sensors (S1, S2, S3, ..., Sn) of the measurement system, wherein in said method... - Provides compressed sensor signal data describing sensor signals, wherein the sensor signal data represents a signal object, the signal object being assigned the signal waveform characteristics of the sensor signal, the signal waveform characteristics being extracted from the sensor signal by means of feature extraction (FE) and forming features of the total feature vector signal (F1). For each signal object, data describing the corresponding signal object is fed to another artificial neural single decompression network (ENN1, ENN2, ..., ENNn) among multiple artificial neural single decompression networks, or a corresponding signal object is generated based on the data describing the signal object and said signal object is fed to another artificial neural single decompression network (ENN1, ENN2, ..., ENNn), wherein said neural single decompression network (ENN1, ENN2, ..., ENNn) is parameterized, and said parameterization is inverse in terms of processing from input signal to output signal compared with the processing from input signal to output signal by artificial neural networks or by other data processing devices used in the compression of said sensor signal data. - Each neural network is decompressed individually (ENN1, ENN2, ..., ENNn) to form a reconstructed single feature vector signal (ER1, ER2, ..., ERn). - The reconstructed individual feature vector signals (ER1, ER2, ..., ERn) are summed to form a reconstructed total feature vector signal (ER) representing the total feature vector signal (F1), which represents the decompression of the compressed sensor signal data.

3. The method according to claim 2, characterized in that, The reconstructed single feature vector signal (ER1, ER2, ..., ERn) is formed by spatial or temporal multiplexing.

4. The method according to claim 2 or 3, characterized in that... ， The method according to claim 1 or 2 provides compressed sensor signal data describing the sensor signal, wherein parameterization of each of the neural single decompression networks (ENN1, ENN2, ..., ENNn) produces a reconstructed single feature vector signal (ER1, ER2, ..., ERn) assigned to the signal object and is the inverse of the neural single reconstruction network (NN1, NN2, ..., NNn) in terms of processing the input signal into the output signal.

5. The method according to claim 1, characterized in that, The detection area is the detection area adjacent to the vehicle.

6. The method according to claim 1, characterized in that, The detection area is the vehicle's surrounding environment.

7. The method according to claim 1, characterized in that, The sensor is an ultrasonic sensor.

8. The method according to claim 1, characterized in that, Potential obstacle objects within the detection area are determined based on the obstacle object feature vector signals (RO1, RO2, ..., ROp) regarding their position, orientation and / or orientation and / or their type, nature and / or distance from the vehicle.

9. The method according to claim 2, characterized in that, The measurement system is a distance measuring system.

10. The method according to claim 2, characterized in that, The sensor is used in vehicles.

11. The method according to claim 2, characterized in that, The compressed sensor signal data describing the sensor signal is the compressed sensor signal data describing the ultrasonic sensor signal.

Images