System for ultrasonic-based object detection

Abstract

The present invention relates to a system (100, 200, 3000,400, 500) for ultrasound-based object detection, comprising: a plurality of ultrasound receivers (106, 110, 404) configured to receive at least two ultrasound echoes of different frequencies; an evaluation unit (104) configured to periodically perform single-channel object detection, where the single-channel object detection involves analyzing the ultrasonic echoes of one of the frequencies to detect the object (102), where, during the single-channel object detection, for a plurality of cycles, the plurality of frequencies are alternately used according to a predetermined scheme, and in the plurality of cycles, the plurality of frequencies are alternately used according to a predetermined scheme. Multi-channel object detection is performed per cycle, where the multi-channel object detection is an analysis of the result of the currently performed single-channel object detection and the result of at least one other past single-channel object detection of the ultrasonic echoes of another of the plurality of frequencies for the purpose of detecting the object, and outputting a result of the multi-channel object detection as a result of the ultrasonic wave-based object detection of the current period.

Description

Technical Field

[0001] This invention relates to a system for evaluating ultrasonic echoes in order to detect objects. Such a system could be used, for example, in motor vehicles to prevent collisions. Background Technology

[0002] In the prior art, ultrasonic measurement systems are known to evaluate the reflection of ultrasonic waves in order to detect moving or stationary objects.

[0003] For example, German patent DE 11 201 8 001 826 B3 describes object detection for vehicles using ultrasonic waves. German patent application DE 102012222891 A1 describes a driver assistance system using ultrasonic sensors. Such systems can be used, for example, in conjunction with parking assistance systems or accident prevention systems.

[0004] Simultaneous operation of multiple sensors to improve spatial resolution is often difficult to achieve or does not yield the desired improvement in object detection because the signal from one sensor can also be measured in adjacent sensors (cross-echo problem).

[0005] In order to make the ultrasonic signals of different ultrasonic sensors distinguishable, German patent specification DE 10 201 7123 049 B3 proposes to modulate the envelope of ultrasonic pulse packets from different transmitters by varying the frequency variation pattern.

[0006] The object of this invention is to provide an improved method and system for ultrasonic-based object detection. This object is achieved through the features of the independent claims. Advantageous embodiments are described in the dependent claims. Summary of the Invention

[0007] In one aspect, an ultrasonic-based object detection system is disclosed. The system includes multiple ultrasonic receivers and an evaluation unit. The ultrasonic receivers are configured to receive ultrasonic echoes of at least two different frequencies. The evaluation unit is configured to periodically perform single-channel object detection. Single-channel object detection is object detection that includes analyzing the ultrasonic echo of one of the frequencies to detect an object, wherein, during single-channel object detection, multiple frequencies are used alternately according to a predetermined scheme for multiple cycles. The evaluation unit is also configured to perform multi-channel object detection in each of the multiple cycles. Multi-channel object detection is object detection for the purpose of detecting an object, including analyzing the result of the currently performed single-channel object detection and the result of at least one other previously performed single-channel object detection in the ultrasonic echo of another of the multiple frequencies. The evaluation unit is also configured to output the result of the multi-channel object detection as the result of the ultrasonic-based object detection for the current cycle.

[0008] Using multiple different frequencies for object detection has the advantage of improving object detection quality because different objects sometimes have different sound reflection characteristics at different frequencies and / or because there are interfering echoes in the environment at certain frequencies that are either absent or significantly weaker at other frequencies. On the other hand, the use of multiple frequencies can also introduce problems, such as cross-echoes or incompatibility with previously used object detection algorithms designed to evaluate echoes at specific frequencies. The applicant has observed that by performing object detection for different ultrasonic echo frequencies in a frequency-specific manner (single-channel object detection), the problems associated with the use of multiple different frequencies can be avoided or at least reduced, where the results of multiple channels are considered when calculating the final result regarding the presence of an object. Therefore, embodiments of the present invention can have the advantage of using multiple different frequencies for object detection without the detection of object “flickering,” particularly at the edges of the sound beam, which causes problems in object detection. The advantages of using multiple different frequencies include, in particular, improved robustness of object detection to false positive object detection results.

[0009] Another advantage of combining multi-channel object detection with multiple single-channel object detection is reduced sensitivity to interference, particularly incorrect object detection errors due to cross-echoes. Such errors, for example, arise in frequency-modulated methods. Furthermore, ultrasonic signals of different frequencies have beam sizes and reflection characteristics, making object detection more difficult, especially at the edges of the corresponding beams, and in some cases, impossible. The embodiments of the systems and methods for object detection described herein do not exhibit these problems or only exhibit them to a reduced degree.

[0010] Frequency-based object detection can be achieved, for example, by having a corresponding transmitter alternately emit ultrasonic sequences of corresponding frequencies in time, and by analyzing the echo signal specifically for the echo of the currently active frequency. Alternatively, a frequency filter can be used to ensure that only echo signals of specific frequencies are considered when performing single-channel object detection. This means that the negative effects of cross-echoes during single-channel object detection can be avoided. The applicant has observed that other problems may arise in object detection due to the different beam characteristics of different signal frequencies. In particular, for objects at the edge of the beam, it is possible that these objects can only be detected by evaluating echo signals of specific frequencies, and not when evaluating echo signals of other frequencies. This leads to “flickering” or contradictory object detection when using multiple different frequencies or channels. However, this problem can be successfully overcome when calculating whether an object has been detected by considering the results of single-channel object detection of two or more channels. Since the channels are used continuously and alternately, multi-channel object detection includes at least one past cycle, or at least one past single-channel object detection. However, in some implementation variations, multi-channel object detection may also include two or more past cycles at a predetermined frequency.

[0011] Therefore, by taking into account the results of one or more object detections in other channels / frequency, it is even possible to reliably detect objects located at the edge of the sound beam at one or more frequencies.

[0012] Preferably, each of the multiple frequencies corresponds to a corresponding channel.

[0013] The system may include, for example, multiple ultrasonic transmitters configured to emit ultrasonic signals, the ultrasonic echoes of which are generated by the reflection of the ultrasonic signals from an object.

[0014] The system may include, for example, multiple ultrasonic transmitters configured to transmit ultrasonic signals at a first frequency and a second frequency of a plurality of frequencies, wherein ultrasonic echoes are generated by the ultrasonic signals due to reflections from an object.

[0015] For example, multiple ultrasonic transmitters can each be configured to emit ultrasonic signals of a single specific frequency from a plurality of different frequencies. For example, an ultrasonic transmitter may include one or more transmitters specifically for emitting ultrasonic signals of a first frequency, and one or more additional transmitters specifically for emitting ultrasonic signals of a second frequency. Optionally, an ultrasonic transmitter may also include one or more additional transmitters specifically for emitting ultrasonic signals of an nth frequency, where n may be, for example, an integer of 3, 4, 5, or higher.

[0016] According to some examples, multiple ultrasonic transmitters are configured to emit ultrasonic signals of different frequencies from multiple different frequencies in a periodic alternation, wherein at any given time, only one of the multiple different frequencies is emitted per cycle.

[0017] This can be advantageous because it avoids cross-echoes, as only one of several frequencies of ultrasonic signal is emitted at any given time. Furthermore, this variant is easier to construct because a dedicated evaluation frequency filter for the echo at a specific frequency can be omitted. These examples are also particularly accurate because even with a frequency filter, problems that may be caused by cross-echoes during object detection cannot be completely eliminated.

[0018] According to other examples, two or more transmitters operate simultaneously, each emitting ultrasonic signals at a different frequency. In this case, the system is configured to use frequency filters in a periodic alternation manner, such that alternating single-channel object detection considers only the ultrasonic echo of one of the frequencies in each cycle and filters out echoes of other frequencies each time.

[0019] Frequency filters can be, for example, so-called "optimal filters" (or matched filters) that detect the presence of a predetermined signal object and / or completely or extensively filter out signals within a specific frequency range. For example, these filters can optimize the signal-to-noise ratio (SNR). In the presence of interference sources (parameter estimation), matched filters are used for the optimal determination of the presence (detection) of a predetermined signal object with a known signal shape. These interference sources can include signals from other ultrasonic transmitters, echoes outside a specific frequency range, and / or ground echoes.

[0020] According to some examples, the system also includes a control unit. The control unit is configured to alternately activate multiple ultrasonic transmitters during multiple cycles to emit ultrasonic signals according to a transmission scheme, such that, preferably at any given time within one cycle, only one of multiple different ultrasonic frequencies is emitted simultaneously. A predetermined scheme for analysis is matched with the transmission scheme.

[0021] For example, the transmission scheme can specify that only a first frequency ultrasonic signal is transmitted for 3 seconds, and during this time, single-channel object detection is performed only based on the ultrasonic echo of the first frequency. Subsequently, only a second frequency ultrasonic signal is transmitted for 3 seconds, and during this period, single-channel object detection is performed only based on the ultrasonic echo of the second frequency. Afterward, again within a 3-second period, only the first frequency ultrasonic signal can be transmitted and analyzed, or, if the system or transmitter supports a third or more (nth) frequencies, only the third (nth) frequency ultrasonic signal can be transmitted for 3 seconds, during which time single-channel object detection can be performed only based on the ultrasonic echo of the third (nth) frequency.

[0022] This can be advantageous because the problems associated with cross-echo in object detection can be completely avoided. Furthermore, no frequency filter is needed to achieve single-channel object detection at a specific frequency.

[0023] According to some examples, multiple predetermined frequencies (e.g., F1, F2, and / or F3) are not constant throughout the transmission cycle. Instead, the ultrasonic signal transmitted during a cycle / period of channel activation at a specific frequency can be frequency modulated, for example by transmitting an ultrasonic pulse train with a rising frequency during the pulse train (so-called linear frequency modulation rise), or by transmitting an ultrasonic pulse train with a falling frequency during the pulse train (linear frequency modulation fall). This also makes it easier for the evaluation unit to distinguish the first and second frequencies from each other and also from interference signals / ground echoes.

[0024] For example, multiple ultrasonic echoes of different frequencies can be frequency modulated, and the modulation modes for different frequencies can be different. The evaluation unit is configured to use the differences in frequency modulation to distinguish echoes belonging to different frequencies. This can be useful, for example, if multiple transmitters emitting ultrasonic signals of different frequencies are activated simultaneously, and if echoes of other frequencies or channels are filtered out during the periodic analysis of only one frequency or one channel's echo signal at any given time.

[0025] According to some examples, multiple different frequencies include at least one first frequency and a second frequency, wherein the first frequency is lower than the second frequency.

[0026] Ultrasonic signals of different frequencies can also differ in other characteristics, such as in the amplitude and / or dimension (spatial dimension) of the ultrasonic beam. For example, an ultrasonic signal of a first frequency may have a beam with a shape different from that of an ultrasonic signal of a second frequency.

[0027] For example, transmitters for ultrasonic signals at a first frequency and ultrasonic signals at a second frequency can be arranged and aligned such that the central axes of the sound beams of the first and second frequency ultrasonic signals are arranged to be completely or approximately aligned (e.g., the distance between the axes is less than 20% of the diameter of the largest sound beam in the aligned sound beams, preferably less than 10%). This has the advantage that the echo signals emitted by the object are emitted at substantially the same angle, which aids in object detection. However, if the echo lobes of different frequencies differ in length and / or width, objects at the edges of these lobes may no longer be well detected on a frequency-wide basis.

[0028] The applicant has observed that lower frequency ultrasonic signals and / or echoes are particularly sensitive, and the presence of an object can often still be detected even if the amplitude of the object's echo is no longer distinguishable from background noise or echoes from the ground with higher frequency ultrasonic signals. The system or evaluation unit is preferably configured to consider the absence of an object in single-channel object detection based on a first (i.e., lower) frequency as a more reliable indication of the object's non-existence than the absence of an object in single-channel object detection based on a second (i.e., higher) frequency.

[0029] Ultrasound is understood here as sound with frequencies above the range of human hearing. In particular, it covers frequencies of 20 kHz and above, especially the frequency range of 20 kHz to 1 GHz.

[0030] Preferably, each of the different frequencies differs from its closest frequency by at least 1 kHz, for example, at least 2 kHz, for example, at least 3 kHz, and particularly at least 4 kHz. Specifically, this frequency difference can be between approximately 5 kHz and 7 kHz, for example, approximately 6 kHz. For example, the first frequency could be 47 kHz, and the second frequency could be 53 kHz.

[0031] In some examples, multiple different frequencies are static constant frequencies. This means that as long as one or more ultrasonic transmitters emit ultrasonic signals at that frequency, the frequency remains essentially constant.

[0032] In other examples, the multiple different frequencies are frequency modulation (FM) frequencies. This means that when one or more ultrasonic transmitters emit ultrasonic signals at this frequency, the frequency changes in a predetermined manner, such as according to a “linear frequency modulation rise” or “linear frequency modulation fall” process. Preferably, the maximum deviation of the modulation frequency from its average value during the signal train (“pulse train”) is 3.5 kHz away from that average value (i.e., from the average frequency minus 3.5 kHz up to the average frequency plus 3.5 kHz). Therefore, the frequency variation in the FM implementation variant is typically less than, for example, the frequency difference (average) of the various frequencies of approximately 6 kHz.

[0033] Using two or more ultrasonic signals of significantly different frequencies can be advantageous because it can significantly increase the sensitivity and accuracy of object detection, as well as the size of the space covered by the sound beam. Some objects reflect ultrasonic signals of certain frequencies better than others. Furthermore, the sound beams of ultrasonic signals at different frequencies often have different dimensions. Therefore, the combined evaluation of multiple echo signals of different frequencies can increase the sensitivity of object detection, the robustness of object detection to interfering signals, and the size of the space in which objects can be detected.

[0034] Based on some examples, single-channel object detection uses an algorithm for detecting objects in ultrasonic echoes, which is independent of the multiple different beam dimensions for each of the multiple different frequencies.

[0035] Using this algorithm can be advantageous because single-channel object detection does not require adaptation to the potentially different dimensions of the acoustic beams of ultrasonic signals at different frequencies. Furthermore, this creates a degree of hardware independence: the same algorithm can therefore be used for single-channel object detection, for example, in different vehicle models that differ in the number and / or level of ultrasonic signal frequencies used. The system can also be upgraded or reconfigured by adding additional ultrasonic transmitters that support the additional frequencies. There is no need to customize the algorithm for the dimensions of the acoustic beam at those additional frequencies. Optionally, adjustments can be made to the configuration parameters used by the algorithm, such as adjusting the minimum height limit for the echo signal amplitude, to interpret echo signals from a specific spatial region as detected objects. However, this can be easily done by modifying the configuration file without rewriting or recompiling the program code.

[0036] According to some examples, the evaluation unit is configured to, in the multi-channel object detection of the currently executed cycle, return a result indicating the presence of an object if an object was detected in the single-channel object detection of those cycles within a minimum number of cycles already executed (analyzing the ultrasonic echo of a first frequency in each cycle), and also return a result if no object was detected in the previous cycle where the ultrasonic echo of a second frequency was analyzed. For example, this minimum number can be 1, 2, or optionally 3 or a higher integer.

[0037] This can be advantageous because the first frequency is lower than the second frequency, and therefore, especially at the edges of the sound beam, it can support object detection algorithms with higher sensitivity than the second frequency signal. Their amplitude at the edges of the sound beam is typically so weak that it is no longer possible or reliable to detect objects solely based on the ultrasonic echo of the second frequency. Since it is essentially sufficient if objects have been reliably detected in the ultrasonic echo of the first frequency, i.e., in at least one or preferably multiple previous cycles, objects at the edges of the sound beam can still be reliably detected, although single-channel object detection based on the first and second frequencies yields contradictory results for at least some objects (object "flickering," especially at the edges of the sound beam, because these are detected in the echo signal of the first frequency but not in the echo signal of the second frequency).

[0038] According to some examples, the evaluation unit is configured to return a result indicating that an object does not exist in the multi-channel object detection of the currently executed cycle if no object was detected in the single-channel object detection of those cycles over the minimum number of cycles already executed (analyzing the ultrasonic echo of the first frequency in each cycle), and also if an object was detected in the previous cycle in which the ultrasonic echo of the second frequency was analyzed.

[0039] For example, the minimum quantity can be 1, 2, or optionally 3 or a higher integer.

[0040] This means that if an object is not detected in either the first frequency echo signal (“F1 channel”) or the second frequency echo signal (“F2 channel”), or if an object is detected only in the second frequency echo signal, the object is considered “non-existent.” Since the second frequency beam is typically smaller than the first lower frequency beam, it is certainly possible for an object to be detected only in the F1 channel if it is located at the edge of the beam. However, it should be impossible for an object to be detected only in the “smaller” F2 channel, making this pattern of multiple single-channel object detection results in both F1 and F2 channels evaluated as an artifact. Therefore, an object detected only in the F2 channel but not in the F1 channel is ultimately interpreted as “non-existent” in the multi-channel object detection process.

[0041] According to some examples, the evaluation unit is configured to perform multi-channel object detection in each cycle, where multi-channel object detection can be viewed as, for example, the classification of cross-channel visibility of objects. However, the classification of cross-channel visibility can also be performed and output as part of multi-channel object detection.

[0042] The result space for classification or multi-channel object detection includes at least one of the following possible results:

[0043] If the object is detected in single-channel object detection in at least n most recently executed cycles (analyzing the ultrasonic echo of the first frequency in each cycle) and at least m most recently executed cycles (analyzing the ultrasonic echo of the second frequency in each cycle), then the object is considered "detected across channels".

[0044] If the object is detected in at least n most recently executed cycles (analyzing the ultrasonic echo at a first frequency in each cycle), but not in at least m most recently executed cycles (analyzing the ultrasonic echo at a second frequency in each cycle), then the result is that the object is "detected by single-channel F1 only".

[0045] If the object is detected in at least m most recently executed cycles (analyzing the ultrasonic echo at the second frequency in each cycle), but not in at least n most recently executed cycles (analyzing the ultrasonic echo at the first frequency in each cycle), then the result is that the object is "detected by single-channel F2 only".

[0046] If, in single-channel object detection during these cycles, the object is not detected in at least n most recently executed cycles (in each cycle, the ultrasonic echo of the first frequency is analyzed), and if, in single-channel object detection during these cycles, the object is also not detected in at least m most recently executed cycles (in each cycle, the ultrasonic echo of the second frequency is analyzed), then the result is that the object is "not detected across channels".

[0047] The parameters m and n can each be integers, such as 1, 2, 3 or higher.

[0048] Based on some examples, the evaluation unit is configured to perform one or more of the following actions depending on the results of multi-channel object detection:

[0049] - Choose one of several available cross-cycle algorithms for spatial tracking of the object;

[0050] - If an object is detected, display the object;

[0051] - Output a warning or alarm signal regarding the presence of an object;

[0052] - Initiate automatic or semi-automatic maneuvers to avoid the object.

[0053] This can be advantageous because, for example, in the case of "flickering" object detection—that is, detecting an object only on one or some of the different frequency channels—a more accurate but computationally complex algorithm can be used than consistent object detection on all channels. Therefore, the processor load can be reduced by the evaluation unit or by other system components performing these or other downstream calculations, since a relatively less computationally intensive algorithm is often sufficient for tracking when an object is consistently detected on all channels. Similarly, if multi-channel object detection detects an object—that is, if analysis based on two or more single-channel object detection results concludes that the object is actually present with sufficient probability—the system can issue an acoustic, visual, or other warning to the driver or other system components. Additionally or alternatively, the system can also initiate driving maneuvers, such as to prevent a collision with the object.

[0054] Based on some examples, the results of single-channel object detection and / or multi-channel object detection are calculated based on the following input data:

[0055] - The amplitude of the received ultrasonic echo at the currently evaluated frequency;

[0056] - An index of the difference between the amplitude of the ultrasonic echo at the currently evaluated frequency and the amplitude of the ultrasonic echo at another frequency, in particular the difference or ratio of amplitude.

[0057] - An indicator of the difference between the amplitude of the ultrasonic echo at the currently evaluated frequency and the amplitude of the currently detected ground echo, specifically the difference or ratio of amplitude.

[0058] For example, the amplitude of a ground echo can be determined from a received echo signal using generally known methods. For instance, the probability that an echo originates from the ground can be based on previous assessments of ground echoes. For this purpose, multiple echoes of different frequencies are collected over a specific time period, assuming no objects are within the sensor's detection range. These echoes are also referred to as reference ground echo signals. A statistical (gamma) distribution is then approximated based on these echoes. By matching the currently received ultrasonic echo to this distribution, the probability that a "new" echo being examined originates from the ground can be determined.

[0059] In the example described herein, the evaluation unit can be configured to represent the amplitude of the currently evaluated echo based on the ground echo signal (i.e., based on a predetermined reference ground signal). Specifically, the amplitude can be represented as a ratio of the maximum values ​​of the distribution, or alternatively, the ground echo probability of the currently received and analyzed echo signal can be calculated.

[0060] Ground echoes are determined according to another example, as described in German patent application DE 10 202 2 116 373.3 (June 30, 2022).

[0061] Because not only are the results of past single-channel object detection, which are essentially binary results (object detected or not detected), but also analog values, such as, in particular, the amplitude of the received ultrasonic echo and / or information about the amplitude's difference from other frequencies and / or from ground echoes, the accuracy of object detection can be improved. Specifically, amplitude levels and / or indices can already be used in single-channel and / or multi-channel object detection to determine whether an object is detectable in the echo signal of the currently evaluated channel, and the results of single-channel object detection can be weighted higher or lower in multi-channel object detection based on the amplitude or index of the echo corresponding to that single-channel object detection. For example, multi-channel object detection can be configured to give higher weight or more consideration to single-channel object detection results if the absolute amplitude is high (e.g., exceeding a certain limit), because this is an indication of a strong, meaningful echo signal and therefore an indication of the presence of an object. Additionally or alternatively, multi-channel object detection can be configured to give higher weight or more consideration to single-channel object detection results if an indicator suggests that the amplitude of a frequency is high relative to the amplitude of another frequency and / or relative to ground echoes (i.e., exceeding a certain limit). This could be an indication of a strong, meaningful echo signal, and therefore an indication of the presence of an object, since the signal is particularly pronounced relative to ground echoes and / or cross-echoes at another frequency.

[0062] Preferably, the evaluation unit is configured to continuously detect and store amplitudes at different frequencies during use, wherein preferably, the amplitudes relative to echoes at other frequencies and / or the indices relative to ground signals are also calculated and stored. Collisions or near-collisions are also preferably recorded and stored, enabling the generation of a continuously improving and expanding training dataset based on the collected data. According to some embodiments, the training dataset is reused to train a predictive model, such as a Markov chain, and then the trained model is used to perform multi-channel object detection based on a sequence of multiple results from single-channel object detection.

[0063] In particular, determining the amplitude ratio or amplitude difference, instead of determining the absolute value of the amplitude or in addition to determining the absolute value of the amplitude, can be advantageous because these relative amplitude characteristic parameters allow for, or even better distinguish between meaningful echoes and meaningless (background) noise, since the amplitude of echo signals from different objects can differ for echoes at a specific frequency. Therefore, it is often necessary to determine relative values ​​(amplitude difference or amplitude ratio).

[0064] In some examples, amplitude difference and / or amplitude ratio are determined for each object identified in the echo (e.g., for echo regions where the amplitude is above the minimum value).

[0065] In some examples, if the amplitude of the echo currently being performed on a single-channel object detection is less than a predetermined limit relative to the amplitude of echoes at one or more other frequencies, the single-channel object detection returns a result indicating that no object was detected.

[0066] In some other examples, if the difference between the amplitude of the echo currently being used for single-channel object detection and the amplitude of one or more other frequency echoes is higher than a predetermined threshold, and the amplitude of the echo currently being used for single-channel object detection is less than the amplitude of one or more other frequency echoes, then the single-channel object detection result returned is no object detected.

[0067] For example, the evaluation unit can be configured to access a configuration file specifying a first minimum amplitude level relative to a first frequency and a second minimum amplitude level relative to a second frequency. If the system and the ultrasonic transmitter support additional frequencies, additional minimum amplitude levels relative to the corresponding additional frequencies can also be defined in the configuration file. During single-channel object detection based on a specific frequency, the evaluation unit reads the predetermined minimum amplitude level for that frequency and compares it to the amplitude of the echo signal at that same frequency. If the amplitude in a specific spatial region of the echo signal exceeds the minimum amplitude level, the evaluation unit determines, as a result of single-channel object detection, that an object is present in the ultrasonic echo. Optionally, the location of the object is also determined, for example, as the range by which the echo amplitude exceeds the minimum amplitude stored for the corresponding frequency. The minimum amplitude can be different for different frequencies. Therefore, the evaluation unit can perform single-channel object detection using the minimum amplitude stored for frequencies actively evaluated. Optionally, single-channel object detection may include further computational steps, such as normalization steps, signal smoothing, or noise reduction steps.

[0068] In multi-channel object detection, the amplitudes of the echo signals at the first and second frequencies can also be used as input. For example, multi-channel object detection can be implemented such that it is performed once per cycle, or once per analysis of a specific frequency channel, wherein in the first part of the multi-channel object detection, single-channel object detection of the currently used frequency channel is performed, and in the second part of the multi-channel object detection, the result of this current single-channel object detection is compared with the results of a previous single-channel object detection process for (at least one) other channel or other channels. In some implementation variations, during multi-channel object detection, the difference or ratio of the amplitudes of the echo signals at the first and second frequencies is calculated to check whether an object can be identified on two (or more, or all) channels, even though the object is received at different amplitudes based on the beam characteristics. Furthermore, it can be determined whether the echo is an object or a low-level ground echo. For example, if the amplitude difference or ratio of the first and second frequencies is very large, this can indicate that the difference between the low-amplitude ultrasonic echo and the ground echo or general background noise is insufficient as a reliable / meaningful signal of the presence or absence of an object. This also applies to echoes with very low absolute amplitudes. For example, in some implementation examples, the evaluation unit can be configured not to evaluate echoes with particularly low amplitude ratios compared to other frequencies or with particularly small absolute amplitudes, or not to perform single-channel object detection based on these echoes. For example, in this case, the echoes could be interpreted as ground echoes unrelated to the presence of an object. For example, metrics regarding amplitude difference can be used by multi-channel object detection to differently weight the results of single-channel object detection. This can be done explicitly using rules or implicitly using machine learning techniques during the training phase of the predictive model.

[0069] According to some examples, in the case of "cross-channel detection" results, spatial tracking of an object involves matching the position of the currently detected object with the position of the object detected in the immediately preceding cycle to determine whether the distance between the position of the object determined in the immediately preceding cycle and the current detected position of the object is below a limit value.

[0070] Therefore, if an object is detected successively in multiple successive cycles or in echo signals at multiple successive evaluation frequencies, the time interval between two object detection events is very small. Even if the object has moved relative to the ultrasonic transmitter and / or moved in an absolute sense during the time between two successive cycles, the possible change in position due to object detection performed in each individual cycle is relatively small. In this case, to spatially track the object over multiple cycles, it is sufficient to determine whether the object's position relative to the position of the object detected in the previous cycle is no longer far from the proposed limit, in which case it is possible to detect that the object detected in the successive cycles is the same object, which may have only changed its absolute or relative position. If the spatial distance exceeds the limit, the evaluation unit assumes that the objects are different, or if the object detection is not continuous, it assumes the presence of interference signals or noise.

[0071] The algorithm described above for tracking objects can be advantageous because it is not very computationally intensive. The simple comparison of the distance between the positions of detected objects in successive cycles with a limiting value is not computationally expensive, thus being particularly advantageous for real-time systems with limited computational power. Such systems are often used specifically in the automotive field.

[0072] According to some examples, the system is designed as a vehicle component and / or part of the vehicle. For example, the ultrasonic transmitter and sensor may be integrated into the front and / or rear bumper, but may also be integrated into the side of the door, the roof, or the floor, for example, to detect obstacles in all three dimensions. Preferably, the limit value of the maximum distance between objects that are considered the same object in two successive cycles depends on the vehicle speed, where the limit value increases with increasing vehicle speed.

[0073] Additionally or alternatively, in the case where the result is "detected only in single-channel F1", spatial tracking of the object includes:

[0074] The future position of an object is predicted by analyzing the position data of the object detected by analyzing the ultrasonic echoes of the first frequency in the current cycle and one or more previously executed cycles.

[0075] In the next cycle of analyzing the ultrasonic signal at the first frequency, the position data of the object obtained by the currently analyzed ultrasonic echo will be matched with the predicted position to determine whether the distance between the predicted position of the object and the current detection position of the object is below the limit value.

[0076] The evaluation unit is configured to treat objects detected in different periods as the same object if the position difference is below a limit value; otherwise, it returns the presence of two different objects or noise signals as the result of the tracking.

[0077] Therefore, if an object is detected only in channel F1 and not in channel F2—for example, if the object blinks between "detected" and "not detected" as it changes channels—the object's future position is predicted by computation (with high complexity), for example, by extrapolating the object's position change across the previous two object detection results based on F1 channels. Since there is a longer time interval between position determinations within an F1 cycle, during which the object can move, predicting the object's position based on its previous position change data ensures that the object can still be reliably tracked. A simple comparison based on the maximum distance limit would be less accurate here because the time interval between the compared position data is too large. Therefore, this computationally complex form of object tracking is selectively used only when there is no reliably continuous detection of an object in each frequency channel. In particular, if three or more different frequencies are used, and the object is detected only in the channel with the lowest frequency, for example, the object's position is detected only in every third cycle (or even less frequently in the case of more frequencies). However, in this case, regardless of the time interval between location determinations, the fact that the object's position can be predicted for the next cycle using the same channel again allows for a sufficiently accurate determination of whether the object is the same object previously detected but merely moving in space, or a different object.

[0078] Based on some examples, the evaluation unit uses a trained prediction model to perform multi-channel object detection.

[0079] This can be advantageous because machine methods are able to detect even complex, and possibly nonlinear, patterns within even longer sequences of single-channel object detection results, reliably indicating whether an object actually exists (or does not exist).

[0080] In contrast, rule-based system detection is typically limited in the complexity of the patterns it can detect.

[0081] In particular, the trained prediction model can be a Markov chain. This form of prediction model has been found to be particularly suitable for reliably detecting the actual presence of objects based on two or more different sequences, possibly even longer sequences of single-channel object detection results.

[0082] Based on some examples, the trained prediction model is a Markov chain that includes the following states:

[0083] - Status "No object detected in multi-channel object detection based on first frequency (F1) or second frequency (F2)";

[0084] - Status "Object detected in multi-channel object detection based only on the first frequency (F1)";

[0085] - Status "Object detected in multi-channel object detection based only on the second frequency (F2)";

[0086] - Status "An object was detected in multi-channel object detection based on both the first frequency (F1) and the second frequency (F2)".

[0087] Therefore, the state corresponds to the possible results of multi-channel object detection (and can contain more states / nodes accordingly, for example, if a third and additional frequencies are used).

[0088] Each state is connected to another state via a path, and the path is assigned a transition probability, which is obtained specifically through statistical analysis of the training data, which includes observed sequences of single-channel object detection results and annotated multi-channel object detection results.

[0089] For example, a training dataset can be created by positioning an object relative to a system for ultrasonic-based object detection, and then a series of single-channel object detection results can be calculated based on echo signals of multiple different frequencies, as already described in several places. Preferably, the object and the system move relative to each other at least in some test runs. For example, the object may move relative to the system (driven, towed, etc.) and / or the system may move relative to the object while receiving and processing ultrasonic echoes of various frequencies. For example, the system may be installed in a vehicle, and the vehicle may move relative to the object at different speeds in different test series. Preferably, the object has different sizes, orientations, shapes, surface and / or material compositions in different test series. Furthermore, multiple series of tests are performed in which no object is positioned within the sound level of the ultrasonic sensor. The training dataset is created based on the knowledge of the test series where the object actually exists, while when it does not exist, the series of single-channel object detection results calculated for different test series and different frequencies are annotated with the knowledge of whether the object actually exists and should be detected. Preferably, each single-channel object detection result also includes annotations about the amplitude of the echo signal evaluated in the single-channel object detection, and / or an index specifying the difference between that amplitude and the amplitudes of echo signals or ground echoes at different frequencies. Single-channel object detection performed to create a training dataset is also referred to as training single-channel object detection. Annotations regarding whether an object actually exists can be done manually, especially when the training dataset is created for the first time. In some embodiments, annotations can be done automatically or semi-automatically, for example, in the form of multi-channel object detection results from an existing prediction model.

[0090] After the training dataset has been created, a prediction learning procedure is performed on the training dataset to obtain a trained prediction model that can be integrated into the evaluation unit and used to perform multi-channel object detection results.

[0091] Specifically, the trained prediction model can be configured and used to compute multi-channel object detection results based on the observed sequence of single-channel object detection. The observed sequence used as input to the model and evaluated can in particular include at least three consecutively obtained results of single-channel object detection, especially at least four, and especially at least five.

[0092] Therefore, a predictive model can be obtained by applying machine learning methods to training data, which includes sequences of observed training single-channel object detection results and annotated training multi-channel object detection results. During training, the model learns to correlate multi-channel object detection results with sequences of single-channel object detection results in order to predict multi-channel object detection results based on sequences of successive single-channel object detection results.

[0093] Preferably, the training data also specifically includes one or more of the following parameter values, which are assigned to one of the training single-channel object detection results in each case, such that during training, the model learns to associate multi-channel object detection results with these parameter values ​​so as to also predict multi-channel object detection results based on these parameter values:

[0094] The amplitude of the ultrasonic echo evaluated during training for single-channel object detection;

[0095] An index of the difference between the amplitude of an ultrasonic echo evaluated in training single-channel object detection and the amplitude of an ultrasonic echo at another frequency of different frequencies, specifically the difference or ratio of amplitudes.

[0096] In training single-channel object detection, the evaluation index is the difference between the amplitude of the ultrasonic echo and the amplitude of the ground echo, specifically the difference or ratio of amplitude.

[0097] This can be advantageous because the model can use amplitude or metrics to identify how reliable or meaningful a single-channel object detection result is within a specific amplitude range. Generally, single-channel object detection results are more reliable if the absolute and / or relative amplitude is larger compared to other echoes (including ground echoes), and are given more consideration in the calculation of multi-channel object detection results.

[0098] Specifically, the prediction model can be a Markov chain. For example, before training begins, the Markov chain can be specified initially without transition probabilities, or the probabilities can be initialized with preset values ​​(e.g., 50% or other estimates). During training, the transition probabilities between the final results of multi-channel object detection can be determined by the fact that the actual observed sequences of these results, identified and included in the training dataset during a period of time using the system or method described herein, are statistically analyzed to calculate the probability of a specific state occurring in the next cycle for a given state in the current execution cycle (based on the next analysis of the next frequency channel of the scheme). The transition probabilities calculated through this analysis can now be integrated into the Markov chain. Furthermore, the model learns to recognize patterns within a series of single-channel object detections in the training data, making the Markov chain generated as a training result usable to predict future states or multi-channel object detection results for future cycles based on a series of past single-channel object detection results.

[0099] On the other hand, a method for ultrasonic-based object detection is disclosed. The method is performed periodically for each of a plurality of different frequencies. For each of the plurality of different frequencies, the following steps are performed, wherein the plurality of frequencies are used in a periodic alternation manner according to a predetermined scheme, and the plurality of steps are repeated for the newly used frequency in a new cycle:

[0100] - Receive ultrasonic echoes of this frequency through one or more ultrasonic receivers;

[0101] - Single-channel object detection is performed by the evaluation unit, whereby single-channel object detection involves analyzing ultrasonic echoes at one frequency to detect objects.

[0102] - Perform multi-channel object detection, where multi-channel object detection is for the purpose of detecting objects, analyze the results of the currently performed single-channel object detection and the results of at least one other previously performed single-channel object detection in the ultrasonic echoes of another of multiple frequencies, and

[0103] - Output the results of multi-channel object detection as the results of ultrasound-based object detection for the current cycle.

[0104] On the other hand, a method for providing an evaluation unit for ultrasound-based object detection is disclosed. The method includes:

[0105] - Provide training data, which includes a sequence of observed training single-channel object detection results and annotated training multi-channel object detection results. Each training single-channel object detection result includes: an indication of whether an object can be detected in an ultrasonic echo at one of several different frequencies; the frequency of the ultrasonic echo analyzed during single-channel object detection; and optionally, a parameter value of the amplitude of the echo and / or an index of the difference between the amplitude and the echo at another frequency or a ground echo. The training data contains a sequence of continuously obtained single-channel object detection results annotated with information about the actual presence of objects.

[0106] - A machine learning method is applied to generate a predictive model, wherein the model learns during training to correlate multi-channel object detection results with a sequence of single-channel object detection results and optionally with parameter values, so as to predict multi-channel object detection results based on a sequence of successive single-channel object detection results at different frequencies, and optionally also based on parameter values.

[0107] - Integrate the predictive model into an evaluation unit for ultrasound-based object recognition; and

[0108] - Provide an evaluation unit, for example, which can be integrated into and used in a system for ultrasonic-based object detection, as illustrated herein with various examples.

[0109] According to an embodiment, the results of ultrasonic-based object detection are used to autonomously or semi-autonomously control the vehicle and / or warn the driver about objects.

[0110] It goes without saying that, as long as the embodiments are not mutually exclusive, one or more of the foregoing embodiments can be combined with each other.

[0111] "Channel" is understood herein as an operating mode of the system in which only ultrasonic echoes of a specific frequency are analyzed by the evaluation unit. This operating mode can be implemented in various ways, for example, by controlling one or more ultrasonic sensors such that they always emit ultrasonic signals at only one of a plurality of predetermined frequencies, with the channel associated with that frequency being active. Alternatively, multiple hardware-based or software-based frequency filters can be controlled or used in such a way that they always receive or analyze ultrasonic echoes at only one of a plurality of predetermined frequencies, with the channel associated with that frequency being active. The period of the method described herein corresponds to the time period during which a specific channel is active.

[0112] When a channel for a specific frequency is "active", the system can be configured such that a transmitter used only to transmit signals at that frequency is active during that time period, and a receiver used only to receive reflections (i.e., echo signals) of that signal is active.

[0113] Here, "control unit" refers to a software and / or hardware module configured to control multiple ultrasonic transmitters. The control unit may be, for example, a component of an evaluation unit or operatively connected to it. Specifically, control can be performed such that the ultrasonic transmitters transmit ultrasonic signals in a periodic, alternating manner according to a transmission scheme, such that, preferably at any given time within a period (which corresponds to the activation time of a channel), only one of multiple different frequencies of ultrasonic signals is transmitted simultaneously. The predetermined scheme for analysis is matched with the transmission scheme.

[0114] The term "evaluation unit" is understood herein to refer to a data processing system configured to evaluate ultrasonic echoes for detecting objects in an environment that includes an ultrasonic sensor. For example, the evaluation unit can be software, hardware, firmware, or a combination thereof. The evaluation unit can be implemented, for example, as an "embedded system" and includes one or more microprocessors, implemented together with the ultrasonic sensor and / or ultrasonic transmitter as a build kit or retrofit kit for a vehicle. However, the evaluation unit can also be a component of central vehicle control logic.

[0115] "Object detection" is understood here as a computer-based evaluation of ultrasonic echoes, in which the presence of one or more objects within a space covered by the original ultrasonic signal is attempted. Optionally, object detection may also include detecting the location of objects and / or other object characteristics (size, shape, object type, etc.). Object detection can be implemented in various ways. In a simple case, object detection involves testing whether the amplitude of the echo signal exceeds a predetermined threshold, where if the threshold is exceeded, the presence of an object is detected. Other implementation variations may use more complex object detection methods, such as Markov chains or neural networks, or other machine learning methods. Combinations of simple and complex amplitude comparison-based algorithms are also possible.

[0116] Here, "object tracking" is understood to refer to a computer-based evaluation of ultrasonic echoes over a period of time (e.g., over multiple periods or frequency channel variations) for the purpose of tracking the absolute and / or relative spatial movement of a detected object. For example, the object can be a moving object. For instance, relative motion can occur between the object and a vehicle containing the system described herein (including ultrasonic sensors).

[0117] A Markov chain is understood as a function describing a stochastic process and used to indicate the probability of future events (also known as states) occurring. Markov chains are based on the assumption that only finite prior history knowledge makes it possible to predict future developments as well as (or at least sufficiently well in the relevant application context) as knowledge of the entire history of the process. In a first-order Markov chain, the future state of the process is calculated solely from the current state and does not depend on any other past states. In the case of an n-order Markov chain, the future state is calculated from n previous states. Attached Figure Description

[0118] The examples are explained in more detail below with reference to the accompanying drawings. In these:

[0119] Figure 1 A system with separate ultrasonic transmitters for different frequencies is shown.

[0120] Figure 2 A system with an ultrasonic transmitter is shown, which alternately emits signals of different frequencies.

[0121] Figure 3 A system with multiple ultrasonic transmitters for each of two different frequencies is shown.

[0122] Figure 4 A system with multiple ultrasonic transmitters for each of three different frequencies is shown.

[0123] Figure 5 A system with an ultrasonic transmitter is shown that alternately emits signals at three different frequencies.

[0124] Figure 6 A block diagram of the evaluation unit is shown.

[0125] Figure 7 A block diagram showing the functions and submodules of the evaluation unit is provided.

[0126] Figure 8 A Markov chain is shown.

[0127] Figure 9 A flowchart of an ultrasonic-based object detection method is shown.

[0128] Figure 10 The matrix showing the single-channel object detection results is illustrated.

[0129] Figure 11 The acoustic cones of two ultrasonic signals at different frequencies are shown. Detailed Implementation

[0130] In the following text, similar elements are labeled with the same reference numerals.

[0131] Figure 1 System 100 is shown, comprising a first ultrasonic receiver 106 for receiving ultrasonic echoes at a first frequency (F1) and a second ultrasonic receiver 110 for receiving ultrasonic echoes at a second frequency (F2). These receivers may be attached to or embedded in a component, such as a vehicle bumper or other component. Receivers 106, 110 are communicatively connected to an evaluation unit 104. The evaluation unit may be designed as part of a data processing system. The evaluation unit, together with sensors and optionally ultrasonic transmitters 108, 112, may constitute a single component and / or may be mounted in or on the same component where ultrasonic receivers are already installed. However, the evaluation unit may also be a separate component. For example, the evaluation unit may be a module of central vehicle control logic and communicatively coupled to the ultrasonic sensors and / or ultrasonic transmitters via a data bus or other wired or wireless communication connection. The system also includes a first ultrasonic transmitter 108 configured to transmit an ultrasonic signal 114 at a first frequency (F1), and a second ultrasonic transmitter 112 configured to transmit an ultrasonic signal 116 at a second frequency (F2). For example, the first frequency may be lower than the second frequency.

[0132] Receiver 106 is therefore designed to receive the echo of signal 114 from transmitter 108, and receiver 110 is designed to receive the echo of signal 116 from transmitter 112. The system can have various mechanisms to ensure that cross-echoes are prevented, i.e., receiver 106 essentially receives only the echo of a first frequency, and receiver 110 essentially receives only the echo of a second frequency. For example, transmitter / receiver pairs 108 / 108 and 110 / 112 can operate alternately in time, such that at any given time, only one of the two frequencies of ultrasonic signal and echo can be transmitted or received. Alternatively, the alternating use of filters that allow only ultrasonic echoes of a specific frequency or a specific frequency range to pass through is possible.

[0133] Even though the reflection of ultrasound waves on objects and / or the Doppler effect may cause slight differences in the frequencies of the original signals 114, 116 and the received echoes, these relatively small frequency shifts can be ignored in the context of the object detection and object tracking applications described herein, such that it is assumed here that the echo signal of the ultrasonic signal 114 has essentially a first frequency F1 and the echo signal of the ultrasonic signal 116 has essentially a second frequency F2.

[0134] The ultrasonic sensor 106 and the ultrasonic transmitter 108 can be implemented as separate components or as a single component (ultrasonic transceiver). Similarly, the ultrasonic sensor 110 and the ultrasonic transmitter 112 can be implemented as separate components or as a single component.

[0135] Figure 2 A system with an ultrasonic transmitter 106 is shown, which alternately transmits signals of different frequencies, for example, according to a predetermined timing scheme. The ultrasonic transmitter can be sensed by an evaluation unit 104 to transmit an ultrasonic signal of a first frequency F1 during a first time period t1. During time t1, an ultrasonic receiver 108 receives the echo of the first frequency and forwards the echo signal to the evaluation unit 104 for evaluation and object detection. During t1, the evaluation unit performs single-channel object detection based on the echo signal of the first frequency and additionally performs multi-channel object detection. According to this scheme, the evaluation unit initiates the transmission of an ultrasonic signal of a second frequency F2 during a long time period t2 after time period t1 has elapsed. During time t2, the ultrasonic receiver 108 receives the echo of the second frequency and forwards the echo signal to the evaluation unit 104 for evaluation and object detection. During t2, the evaluation unit performs single-channel object detection based on the echo signal of the second frequency and additionally performs multi-channel object detection. After time period t2 expires, the process can be restarted from the beginning (during time period t1, ultrasonic signals or echoes of the first frequency are sent and evaluated).

[0136] Figure 3 A system is shown having a plurality of ultrasonic transmitters 108.1, 108.2 for transmitting an ultrasonic signal 114 at a first frequency and a plurality of other ultrasonic transmitters 112.1, 112.2 for transmitting an ultrasonic signal 116 at a second frequency. The system also includes a plurality of receivers 106.01, 106.2 for receiving ultrasonic echoes at the first frequency, and additional receivers 110.1, 110.2 for receiving ultrasonic echoes at the second frequency.

[0137] Figure 3 The transmitter and receiver shown can be controlled, for example, as Figure 1 The description of the corresponding transmitters and receivers includes, for example, all transmitters of an ultrasonic signal at a first frequency simultaneously transmitting that frequency, and all transmitters of an ultrasonic signal at a second frequency simultaneously transmitting that second frequency. In other implementation variations, a significantly larger number of transmitters and receivers may exist for each frequency, for example, 4, 5, 6, 7, or more transmitters and / or receivers in each case. By using multiple transmitters and receivers for each frequency, the spatial resolution and accuracy of object detection can be improved.

[0138] Figures 3 to 4 The multiple sensors shown can be designed as components, for example, of an ultrasonic transceiver. Each sensor typically receives not only the echo signal from the ultrasonic transmitter of its respective transceiver, but also echoes from ultrasonic signals from other transceivers at the same frequency. By appropriately evaluating the echo signals generated by multiple transmitters on the same object, each echo signal is detected by a receiver, the object's position can be accurately determined.

[0139] Figure 4 A block diagram of the system is shown, which has two ultrasonic transmitters and two ultrasonic receivers for each of three different frequencies F1, F2, and F3. For example, ultrasonic signal 114 may have the lowest frequency F1 (and the largest beam), ultrasonic signal 116 may have a higher frequency F2 (and a medium beam), and ultrasonic signal 402 may have the highest frequency (and the smallest beam).

[0140] For example, ultrasonic transmitters of different frequencies can be spaced approximately 5 cm to 50 cm apart from the nearest transmitter (of the same or another frequency), particularly between 10 cm and 40 cm, and especially between 15 cm and 35 cm. The sound beams of the transmitters can be several meters long, typically greater than 1 m, particularly greater than 2 m, and in some cases even greater than 4 m, 10 m, or even greater than 20 m. Typically, the length of the sound beam covers a space up to 5 m from the corresponding ultrasonic transmitter. Compared to this length, the distance between transmitters is relatively small, meaning that at least some of the sound beams from different transmitters can strongly overlap and, in some cases, extend approximately along the same axis (see [link to relevant documentation]). Figure 11 In particular, at the edges of overlapping sound beams, discontinuous "flickering" object detection may occur, i.e., an object is detected in one or some channels but not in one or more other channels.

[0141] For example, during a first time period corresponding to the first cycle, only transmitters 108.1 and 108.2 can be activated to emit ultrasonic signals at a first frequency, while the other transmitters do not emit any signals. During a second time period corresponding to the second cycle following the first time period, only transmitters 112.1 and 112.2 are activated to emit ultrasonic signals at a second frequency, while the other transmitters do not emit any signals. During a third time period corresponding to the third cycle following the second time period, only transmitters 406.1 and 406.2 are activated to emit ultrasonic signals at a third frequency, while the other transmitters do not emit any signals.

[0142] Figure 5A system with an ultrasonic transmitter is shown that alternately emits signals of three different frequencies. During a time period t1 corresponding to a first evaluation cycle, transmitter 106 emits an ultrasonic signal 114 at a first frequency F1. The echo of this ultrasonic signal is received by receiver 108 and forwarded to an evaluation unit to perform object detection. During a time period t2 following t1 corresponding to a second evaluation cycle, transmitter 106 emits an ultrasonic signal 116 at a second frequency F2. The echo of this signal is received by receiver 108 and forwarded to the evaluation unit to perform object detection in the second cycle. During a time period t3 following t2 corresponding to a third cycle, transmitter 106 emits an ultrasonic signal 402 at a third frequency F3. The echo of this signal is received by receiver 108 and forwarded to the evaluation unit to perform object detection. The cycle sequence can then be restarted from the beginning, and the transmitter can emit an ultrasonic signal of the first frequency during another time interval t1.

[0143] Figure 6 A block diagram of a data processing system 104 is shown, which is designed as an evaluation unit 104 or includes an evaluation unit 104. For example, the data processing system may be a microcontroller or computer having one or more processors 602 and volatile or non-volatile memory 604. Software programs stored in the memory evaluate echo signals received from ultrasonic receivers 106, 110, 404 in order to detect and track objects, i.e., determine the position of the detected objects over time.

[0144] The evaluation unit can be, for example, a component of the vehicle's central control logic or a separate component, which can optionally transmit the object detection results to the central control logic, allowing the latter to initiate braking and evasive maneuvers or issue a warning message to the driver. Software 600 checks the echo signal for certain signal waveform characteristics, i.e., for example, determining whether the amplitude of the ultrasonic echo exceeds a certain limit. The object detection task can be designed such that the ultrasonic systems 100, 200, 300 and their evaluation unit 104 perform approximate object detection based primarily on echo amplitude limitations, such as single-channel and multi-channel object detection, and the results of this object detection are transmitted to the vehicle's central control logic for further processing.

[0145] In a variant implementation, the ultrasonic sensor and receiver are integrated as a sensor arrangement including an evaluation unit 104, which functions as a slave / satellite unit and is interoperable with central vehicle control logic, which functions as a master / central unit. The evaluation unit is connected to the central unit via a bus.

[0146] Figure 7A block diagram illustrating the function and submodules of the evaluation unit 104, purportedly a variant of the implementation, is shown. The evaluation unit is configured to periodically perform object detection in exactly one of a plurality of ultrasonic echoes, wherein the frequency of the currently analyzed echo varies periodically according to a predetermined scheme 700. For example, Figure 7 The exemplary scheme shown can specify that: in the first cycle, only the echo of the first frequency (F1) is analyzed. These echo signals are also referred to as the "F1 channel" or "F1 channel signal". In the second cycle, only the echo of the second frequency (F2) is analyzed. These echo signals are also referred to as the "F2 channel" or "F2 channel signal". In the third cycle, only the echo of the third frequency (F3) is analyzed. These echo signals are also referred to as the "F3 channel" or "F3 channel signal". Various mechanisms exist that allow the system to ensure that only one of multiple frequencies is analyzed at any given time (see...). Figure 1-5 (Description as provided). For example, the evaluation unit can be configured to select different transmitters at different frequencies according to scheme 700 and initiate the transmission of the corresponding ultrasonic signals. Alternatively, the evaluation unit can control the use of frequency filters in a periodic mode, such that only one frequency echo is forwarded to the evaluation unit in the current cycle and used there for object detection. Therefore, the evaluation unit knows which frequency the currently received echo signal has in the currently executed cycle, and whether this has a higher sensitivity for object detection than other cycles corresponding to other higher frequency echoes.

[0147] Therefore, although the echoes analyzed in different cycles have different frequencies and typically different dimensions of the sound beam, the evaluation unit 104 preferably uses a general algorithm 702 for object detection, which is not customized for the expected dimensions of the sound beam. However, this may result in objects being detected only in a "flickering" manner during alternating evaluation cycles, particularly objects at the edge of the sound beam or objects that only transmit strong echo signals at specific frequencies. By means of a multi-channel object detection 704 that also considers one or more results of previous single-channel object detection, objects detected only in a "flickering" manner in one of multiple channels can be reliably detected and tracked. The multi-channel object detection 704 may, for example, include multiple steps, the first step being single-channel object detection based on the ultrasonic echoes of the currently actively analyzed channel. The multi-channel object detection can be implemented as, for example, a rule-based algorithm for evaluating the result matrix, such as... Figure 10 and Figure 11 As shown. Depending on some implementation variations, such as... Figure 8 As shown, Markov chains can be used to compute multi-channel object detection results based on a sequence of previously determined single-channel object detection results.

[0148] Figure 8This is a diagram of a Markov chain 800 that can be used for multi-channel object detection. The Markov chain includes multiple states 806-812, for example:

[0149] - Status 806 indicates the result of multi-channel object detection: "No object was detected in multi-channel object detection based on the first frequency (F1) or the second frequency (F2)";

[0150] - Status 808 indicates the result of multi-channel object detection: "Objects were detected in multi-channel object detection based only on the first frequency (F1)";

[0151] - Status 810 indicates the result of multi-channel object detection: "Objects were detected in multi-channel object detection based only on the second frequency (F2)";

[0152] - State 812 indicates the result of multi-channel object detection: "An object was detected in multi-channel object detection based on the first frequency (F1) and the second frequency (F2)".

[0153] Each of these states 806-812 is, in turn, the result of applying the prediction model to a sequence of successive single-channel object detections at different alternating frequencies. In the case of two different frequencies, such a sequence might look like this, for example:

[0154] Sequence 1: F1+|F2+|F1+|F2+|F1+|F2+|F1+|F2+|F1+|F2+|F1+|F2+|F2+|

[0155] Sequence 2: F1+|F2-|F1+|F2-|F1+|F2-|F1+|F2-|F1+|F2-|F1+|F2-|

[0156] Sequence 3: F1-|F2+|F1-|F2+|F1-|F2+|F1-|F2+|F1-|F2+|F1-|F2+|

[0157] Sequence 4: F1-|F2-|F1-|F2-|F1-|F2-|F1-|F2-|F1-|F2-|F1-|F2-|

[0158] In sequence (or “series”) 1, an object is detected in both channels; in sequence 4, no object is detected in either channel; in sequence 2, an object is detected only in F1; and in sequence 3, an object is detected only in F2. Sequences 2 and 3 thus represent “flickering” detection. In reality, sequences are not always so clearly defined. Typically, even in relatively stable patterns, there will be transitions between these “idealized” four frequency types in at least some segments, and irregularities and “outliers” will exist. During machine learning, the model learns the extent to which the sequence can deviate from the frequencies 1-4 outlined above in order to provide corresponding multi-channel object detection results. Preferably, information about absolute or relative amplitude is also input into the training model or the results calculated by the training model. This absolute or relative amplitude information can be assigned as annotations in the training data to the results of single-channel object detection and provided as input to the evaluation unit in both single-channel and multi-channel object detection.

[0159] In some examples, states in a Markov chain are connected by edges (“paths”), where transition probabilities are assigned to the edges. Transition probabilities can be obtained, for example, by analyzing historical data of single-channel and multi-channel object detection results across a large number of successively executed cycles, which can be provided, for example, as a training dataset. For example, the transition probability via the path pointed to by the arrow from node 806 to 808 indicates the probability that, starting from a state where an object was not detected in the currently analyzed channel or in a previously analyzed channel, an object was detected in the next cycle based on the echo signal at frequency F1, where, for example, the results of the last n single-channel object detections can be assigned to the corresponding states, for example, as feature vectors. For example, a state representing a single-channel object detection result is determined by comparing the echo amplitude to a limit value: if the echo signal in a spatial region has an amplitude higher than a predetermined minimum value, the state “Object detected (in the current channel)” is returned as the result of object detection.

[0160] State 812 means that an object is detected based on the ultrasonic echo of a first lower frequency F1 and a second higher frequency (continuous or stable object detection in multiple channels). State 808 means that an object is detected based on the ultrasonic echo of the first lower frequency F1 but not based on the higher frequency F2 ("flickering" object detection). State 802 means that no object is detected based on the ultrasonic echo of the first frequency F1 or based on the second frequency F2 (stable, no object detected).

[0161] According to this scheme, each transition from one state to another corresponds to a new cycle or the analysis of the echo of another state in a predetermined frequency.

[0162] Figure 8The Markov chain shown can be used to model systems supporting two different frequencies, F1 and F2. If additional frequencies are used, the number of states and edges expands accordingly and is supplemented by the corresponding transition probabilities.

[0163] Figure 9 A flowchart of an ultrasonic-based object detection method is shown. This method can, for example, be used to provide parking assistance for vehicles to support various driver assistance systems designed to achieve, for example, autonomous, partially autonomous, or assisted driving. For instance, if the current or predicted distance between the vehicle and an object is below a minimum distance, a warning signal can be automatically issued to the driver and / or braking or evasive maneuvers can be initiated.

[0164] The method is repeated periodically, wherein in each cycle, the ultrasonic echo of only one of a plurality of predetermined frequencies is evaluated, or wherein only one frequency-specific channel is always active in each cycle. The alternation of cycles between the plurality of frequencies occurs according to a predetermined scheme.

[0165] In step 902, one or more ultrasonic receivers receive ultrasonic echoes at the currently active / active frequency. Thus, for example, in the first cycle, ultrasonic echoes at the first frequency may first be received by one or more first ultrasonic receivers 106 and forwarded to the evaluation unit 104.

[0166] In step 904, the evaluation unit 104 performs single-channel object detection based on the echo received in step 902 in order to detect object 102.

[0167] The evaluation unit then performs multi-channel object detection in step 906. During multi-channel object detection, for the purpose of detecting objects, the evaluation unit analyzes the results of the currently performed single-channel object detection as well as the results of at least one other previously performed single-channel object detection in the ultrasonic echo of another of multiple frequencies.

[0168] For example, single-channel and multi-channel object detection processes can be combined, for instance, based on Markov chains. Alternatively, the results of single-channel object detection at multiple periods and frequencies can be initially stored so that they can be analyzed using rules to return a multi-channel object detection result that integrates the results of multiple single-channel object detection at multiple periods and frequencies.

[0169] In step 908, the result of the multi-channel object detection is returned as the result of the ultrasound-based object detection for the current cycle. This can, for example, be used to dynamically select an object tracking algorithm based on the result of the multi-channel object detection. For instance, if the multi-channel object detection result shows that the object is continuously detected in multiple cycles and at multiple or all predetermined frequencies, a computationally less complex algorithm (e.g., based on a distance limit) can be used to track the object's changes over time. However, if the multi-channel object detection result shows that the object is detected, for example, only in a "flickering" manner, such as only in channel F1 and not in channel F2, a computationally complex algorithm for object tracking can be executed. This includes, for example, predicting the object's position in future cycles and comparing the distance between the measured and predicted object positions with a distance limit.

[0170] Then, in the second cycle, the ultrasonic echo at the second frequency F2 can be received by one or more second ultrasonic receivers 110 and forwarded to the evaluation unit, and steps 902-908 are repeated based on the echo at the second frequency. Thereafter, depending on the number of supported frequencies, the process can start again at frequency F1; otherwise, a third, fourth, or Nth frequency cycle is performed according to a predetermined scheme.

[0171] Variations in frequency or channel period can include the use of different transmitter or filter periods, for example, as referenced. Figure 1-5 As described.

[0172] Figure 10 A matrix is ​​shown representing single-channel object detection results, such as those observed in a variant of the implementation. On the right-hand side, in the first test run (“Series A”), a system comprising multiple first ultrasonic transceivers for a first frequency F1 and multiple second ultrasonic transceivers for a higher second frequency F2 was used to detect the presence of an object near the transceivers in a series of six cycles, wherein the first channel / first frequency is activated or evaluated in each of cycles 1, 3, and 5, and the second channel / second frequency is activated or evaluated in each of cycles 2, 4, and 6.

[0173] In the first series of tests (“Series A”), an object is placed at the center of a space covered by sound beams that largely overlap on two separate channels of the first and second transceivers. The object is reliably detected in both channels (in Series A, the single-channel object detection result is always “yes”). The result for multi-channel object detection in this series of tests is correspondingly “object detected consistently in F1 and F2”. The object is considered to be detected or present.

[0174] In the second series of tests (“Series B”), the object is positioned just outside the sound beam at the second frequency, but still within the sound beam at the first frequency, for example, as shown in the image. Figure 11 As shown. Objects are reliably detected in the F1 channel (in series B, single-channel object detection results are almost always "yes" in the F1 channel and always "no" in the F2 channel). The multi-channel object detection results for this test series are "object flashing / object detected only in the F1 channel, but not in the F2 channel." Objects are considered detected or present, but a computationally more complex algorithm can optionally be used for object tracking compared to test series A.

[0175] The third series of tests (“Series C”) is a situation where artifacts should not occur, at least not in the following circumstances. Figure 11 This is the configuration shown. Because the sound beam of F1 is larger than that of F2 (assuming the object reflects both frequencies equally), the pattern shown in Series C should not occur in reality and is interpreted here as meaning the object is not present, and there may be a malfunction, interference signal, or any other error in object detection. The result of multi-channel object detection for this test series is correspondingly "No object detected".

[0176] In the fourth series of tests (“Series D”), the object was positioned just outside the sound beams at both frequencies. No object was detected in any of the channels (consistently “No” in channels F1 and F2 of Series D). The result for multi-channel object detection in this series of tests was “No object detected”.

[0177] Furthermore, if an object is detected more than once in at least one of the channels, but cannot be tracked within a predetermined maximum number of subsequent periods, the evaluation unit can be configured to return "noise" as the result. For example, if an object detected in the first period is predicted to be at position P2 in the second period, position P3 in the third period, and position P4 in the fourth period, this can occur, for example, if no object is located at or sufficiently close to that position in more than two predicted future positions in the corresponding future periods. For example, the evaluation unit may include a counter that counts the number of consecutive periods in which previously detected objects are no longer detected at their predicted positions, and if this number exceeds a predetermined maximum, returns "noise" and / or "no object detected" as the multi-channel object detection result.

[0178] Figure 11The acoustic cones of two ultrasonic signals at different frequencies are shown. In the example shown here, the acoustic cone 950 of the first, lower frequency ultrasonic signal 114 (F1) is greater than the acoustic cone 952 of the second, higher frequency ultrasonic signal 116 (F2). If the object 102 to be detected, such as a physical object or person, is located at the edge or outside of the F2 acoustic cone 952, as shown here, but still within the F1 acoustic cone 950, this can result in “flickering” object detection in both F1 and F2 channels. A similar situation may occur if the material strongly reflects the first frequency ultrasonic signal (i.e., generates an echo) but does not reflect the second frequency ultrasonic signal.

[0179] In the example described here, for simplicity, it is assumed that it essentially involves the detection of an object within the range of the received echo signal. For example, in single-channel object detection, if the amplitude of the echo in a spatial region is higher than a predetermined threshold, an object can be considered detected. However, it is also possible to detect multiple objects in the echo signal. Object detection may include (at least approximately) determining the location of one or more objects. For example, known methods from the prior art can be used to determine the location of the object generating the echo from echoes received from one or more ultrasonic sensors.

[0180] Although the invention has been detailed and described in the accompanying drawings and the description above, such description is to be considered exemplary and not restrictive; the invention is not limited to the disclosed embodiments.

[0181] List of reference numerals

[0182] 100 System

[0183] 102 objects

[0184] 104 Evaluation Units

[0185] 106 Ultrasonic Receiver

[0186] 108 Ultrasonic Transmitter

[0187] 110 Ultrasonic Receiver

[0188] 112 Ultrasonic transmitter

[0189] 114 frequency F1 ultrasonic signal

[0190] 116 frequency F2 ultrasonic signal

[0191] 200 system

[0192] 300 system

[0193] 400 system

[0194] 402 frequency F3 ultrasonic signal

[0195] 404 Ultrasonic Receiver

[0196] 406 Ultrasonic Transmitter

[0197] 500 system

[0198] 600 Evaluation Software

[0199] 602 (multiple) processors

[0200] 604 memory

[0201] 700-channel usage plan

[0202] 702 General Object Detection Algorithm

[0203] 704 Multi-channel Object Detection

[0204] 706 Markov Chain

[0205] 800 Markov chains

[0206] 806 status

[0207] 808 status

[0208] 810 Status

[0209] 812 Status

[0210] Steps 902-908

[0211] 920 Multi-channel Object Detection

[0212] 922 Single-channel object detection results

[0213] 950 The edge of the sound beam at the first frequency F1

[0214] 952 The edge of the sound beam at the first frequency F2

Claims

1. A system (100, 200, 3000, 400, 500) for ultrasonic-based object detection, comprising: Multiple ultrasonic receivers (106, 110, 404) are configured to receive ultrasonic echoes of at least two different frequencies. Evaluation unit (104), which is configured as follows: • Periodically perform single-channel object detection, the single-channel object detection including analyzing the ultrasonic echo of one of the frequencies to detect an object (102), wherein, during the single-channel object detection, the plurality of frequencies (700) are used alternately according to a predetermined scheme for multiple periods; and • In multiple cycles, each cycle performs multi-channel object detection, wherein multi-channel object detection is performed for the purpose of detecting the object, analyzing the result of the currently performed single-channel object detection and the result of at least one other previously performed single-channel object detection in another ultrasonic echo at multiple frequencies, and outputting the multi-channel object detection result as the ultrasonic-based object detection result for the current cycle.

2. The system of claim 1 further includes a plurality of ultrasonic transmitters (108, 112, 406), said plurality of ultrasonic transmitters (108, 112, 406) being configured to transmit ultrasonic signals (114, 116, 402) at a first frequency and a second frequency of said plurality of frequencies, wherein, The ultrasonic echo is generated by the ultrasonic signal due to reflection from the object.

3. The system according to claim 2, wherein, Each of the plurality of ultrasonic transmitters is configured to emit an ultrasonic signal at a single specific frequency among the plurality of different frequencies.

4. The system according to claim 2 or 3, wherein, The plurality of ultrasonic transmitters are configured to emit ultrasonic signals of different frequencies of the plurality of different frequencies in a periodic alternation, wherein at any given time, only one frequency of the plurality of different frequencies is emitted in each period.

5. The system according to any one of claims 2 to 4, further comprising a control unit, in, The control unit is configured to alternately activate the plurality of ultrasonic transmitters during the plurality of cycles to transmit ultrasonic signals according to a transmission scheme, such that, preferably at any given time, only one of the plurality of different frequencies of ultrasonic signals is transmitted simultaneously within one of the cycles. The predetermined scheme used for analysis is matched with the launch scheme.

6. The system according to any one of the preceding claims, wherein, The plurality of different frequencies include at least one first frequency and a second frequency, wherein the first frequency is lower than the second frequency.

7. The system according to any one of the preceding claims, wherein, The single-channel object detection uses an algorithm for detecting objects (102) in ultrasonic echoes, the algorithm being independent of the beam dimension of the plurality of different frequencies for each of the plurality of different frequencies.

8. The system according to any one of claims 6 to 7, wherein, The evaluation unit is configured to: in the multi-channel object detection of the currently executed cycle, if the object is detected in the single-channel object detection of the cycles in which the first frequency ultrasonic echo is analyzed within the minimum number of cycles already executed, wherein the object is detected in each cycle, and if the object is not detected in the previous cycle in which the second frequency ultrasonic echo is analyzed, the same result is also returned.

9. The system according to any one of claims 6 to 8, wherein, The evaluation unit is configured to: in the multi-channel object detection of the currently executed cycle, if the object is not detected in the single-channel object detection of the cycles in which the ultrasonic echo of the first frequency is analyzed within the minimum number of cycles already executed, and if the object was detected in the previous cycle in which the ultrasonic echo of the second frequency was analyzed, then the result of the object not being detected is also returned.

10. The system according to any one of the preceding claims, wherein, The evaluation unit is configured to perform one or more of the following actions based on the results of the multi-channel object detection: Choose one of several available cross-period algorithms for spatial tracking of the object; If the object has been detected, then display the object; Output a warning or alarm signal regarding the presence of the object; Initiate automatic or semi-automatic maneuvers to avoid the object.

11. The system according to any one of the preceding claims, wherein, The single-channel object detection is performed based on the following input data: The amplitude of the received ultrasonic echo at the currently evaluated frequency; An index of the difference between the amplitude of an ultrasonic echo at a currently evaluated frequency and the amplitude of an ultrasonic echo at another frequency, wherein the index is particularly the difference or ratio of amplitudes. An index of the difference between the amplitude of the ultrasonic echo at the currently evaluated frequency and the amplitude of the currently detected ground echo, wherein the index is particularly the difference or ratio of amplitude.

12. The system according to claim 10 or 11, wherein, When the multi-channel object detection result is "detected across channels", the spatial tracking of the object includes matching the position of the currently detected object with the position of the object detected in the immediately preceding cycle to determine whether the distance between the position of the object determined in the immediately preceding cycle and the current detected position of the object is lower than a limit value. and / or When the multi-channel object detection result is "only single-channel F1 detected", the spatial tracking of the object includes: • The future position of the object is predicted by analyzing the position data of the object detected by analyzing ultrasonic echoes of the first frequency in the current cycle and one or more previously executed cycles. • In the next cycle of analyzing the ultrasonic signal at the first frequency, the position data of the object obtained using the currently analyzed ultrasonic echo is matched with the predicted position to determine whether the distance between the predicted position of the object and the current detection position of the object is below a limit value. The evaluation unit is configured to: if the position difference is lower than the limit value, then treat the objects detected in the different periods as the same object; otherwise, return the presence of two different objects or noise signals as the result of the tracking.

13. The system according to any one of the preceding claims, wherein, The evaluation unit uses a trained prediction model to perform the multi-channel object detection, the trained prediction model being, in particular, a Markov chain (706, 800).

14. The system according to claim 13, wherein, The trained prediction model is a Markov chain (706, 800), which includes the following states: The status "No object was detected in multi-channel object detection based on the first frequency (F1) or the second frequency (F2)" (806); The state "Objects were detected in multi-channel object detection based only on the first frequency (F1)" (808); The state "Objects were detected in multi-channel object detection based only on the second frequency (F2)" (810); The state "An object was detected in multi-channel object detection based on both the first frequency (F1) and the second frequency (F2)" (812); Each of the states is connected to another state via a path, and the path is assigned a transition probability, which is obtained specifically through statistical analysis of training data, wherein the training data includes observed single-channel object detection result sequences and annotated multi-channel object detection results.

15. The system according to claim 13 or 14, wherein, The trained prediction model is configured to compute the results of the multi-channel object detection based on the observed sequence of single-channel object detection, the observed sequence specifically including at least three consecutively obtained results from the single-channel object detection results, specifically at least four, and specifically at least five. The system of claim 14, wherein the predictive model is obtained by applying a machine learning method to training data, wherein the training data includes a sequence of observed training single-channel object detection results and annotated training multi-channel object detection results, and the model has learned during training to associate the multi-channel object detection results with the sequence of single-channel object detection results in order to predict the multi-channel object detection results based on the sequence of successive single-channel object detection results of various sequences. The training data specifically includes one or more of the following parameter values, which are assigned to one of the training single-channel object detection results in each case, wherein the model has learned during training to associate multi-channel object detection results with these parameter values ​​in order to predict multi-channel object detection results based on these parameter values: o The amplitude of the ultrasonic echo evaluated during training for single-channel object detection; o An index of the difference between the amplitude of an ultrasonic echo evaluated in training single-channel object detection and the amplitude of the ultrasonic echo at another frequency of different frequencies, the index being particularly the difference or ratio of amplitudes. o An index of the difference between the amplitude of the ultrasonic echo and the amplitude of the ground echo, as evaluated in training single-channel object detection, wherein the index is particularly the difference or ratio of amplitude.

16. The system according to any one of the preceding claims, wherein, The plurality of ultrasonic echoes of different frequencies are frequency modulated, and the modulation modes of the different frequencies are different, wherein the evaluation unit is configured to use the difference in frequency modulation to distinguish echoes belonging to the different frequencies, wherein, in particular, an echo of one of the frequencies corresponds to an ultrasonic signal emitted at a rising frequency during the pulse train (linear frequency modulation rising), and an echo of another of the frequencies corresponds to an ultrasonic signal emitted at a falling frequency during the pulse train (linear frequency modulation falling).

17. A method for ultrasonic-based object detection, comprising: For each of a plurality of different frequencies, the following method is performed, wherein the plurality of frequencies are used in a periodic alternation manner according to a predetermined scheme, and the method is repeated for the newly used frequency in a new cycle: The ultrasonic echo of the frequency of (902) is received by one or more ultrasonic receivers (106, 110, 404); The evaluation unit (104) performs (902) single-channel object detection, wherein the single-channel object detection includes the analysis of an ultrasonic echo of one frequency used to detect the object (102). The evaluation unit performs multi-channel object detection in each of the plurality of said cycles, wherein the multi-channel object detection is for the purpose of detecting the object, and analyzes the result of the currently performed single-channel object detection and the result of at least one other previously performed single-channel object detection in the ultrasonic echo of another of the plurality of frequencies, and The results of the multi-channel object detection are output as the results of the ultrasound-based object detection for the current cycle.

18. A method for providing an evaluation unit for ultrasound-based object detection, comprising: Training data is provided, comprising a sequence of observed training single-channel object detection results and annotated training multi-channel object detection results. Each training single-channel object detection result includes: an indication of whether an object can be detected in an ultrasonic echo at one of a plurality of different frequencies; the frequency of the ultrasonic echo analyzed during single-channel object detection; and optionally, a parameter value of the amplitude of the echo and / or an index of the difference between the amplitude and an echo or ground echo at another of the frequencies. The training data comprises a sequence of continuously obtained single-channel object detection results, the sequence being annotated with information about the actual presence of the object. A machine learning method is applied to generate a predictive model, wherein the model learns during training to correlate multi-channel object detection results with a sequence of single-channel object detection results and optionally with parameter values, so as to predict multi-channel object detection results based on a sequence of successive single-channel object detection results at different frequencies, and optionally also based on parameter values. The prediction model is integrated into an evaluation unit for ultrasound-based object detection; and The evaluation unit is provided.

Images