Improved radar detection using previously tracked object information

Abstract

A system includes a processor and a non-transitory computer readable medium storing machine instructions that cause the processor to perform a first fast Fourier transform (FFT) on received radar data to obtain a distance-antenna data array, and performing a second FFT on the distance-antenna data array to obtain a distance-Doppler-antenna data cube. The processor performs peak detection on the distance-Doppler-antenna data cube based on object information to obtain a subset including acknowledgement peaks and candidate peaks. The processor performs angle of arrival calculations for the subset of the range-Doppler-antenna data cubes, and filters candidate peaks in the subset to obtain a point cloud representing an environment. In some embodiments, the acknowledgement peak is detected based on a default threshold, the candidate peak is detected based on at least one adapted threshold, and the at least one adapted threshold is determined based on the object information.

Description

Technical Field

[0001] The present invention relates to an improved radar detection system and method using previously tracked object information. Background Art

[0002] Some radar systems are included in vehicles as part of automated driving assistance systems and are used to assist in sensing the environment around the vehicle. To accurately represent the environment, the radar system performs peak detection to identify objects. However, performing peak detection based on a constant detection threshold within each range bin may result in false negative detections where the signal strength of the radar reflection from an object is close to the noise floor. Summary of the invention

[0003] A system includes at least one processor and at least one non-transitory computer-readable medium storing machine instructions. The machine instructions, when executed by the at least one processor, cause the processor to perform a first FFT on received radar data to obtain a range-antenna data array, and to perform a second FFT on the range-antenna data array to obtain a range-Doppler-antenna data cube. The processor performs peak detection on the range-Doppler-antenna data cube based on object information to obtain a subset of the range-Doppler-antenna data cube. The subset of the range-Doppler-antenna data cube includes confirmed peaks and candidate peaks. The processor performs an arrival angle calculation for the subset of the range-Doppler-antenna data cube, and filters the candidate peaks in the subset of the range-Doppler-antenna data cube to obtain a point cloud representing an environment.

[0004] In some embodiments, confirmation peaks are detected based on a default threshold value, and candidate peaks are detected based on at least one adapted threshold value. At least one adapted threshold value is determined based on object information. In some embodiments, machine instructions for performing peak detection on a range-Doppler-antenna data cube include machine instructions for preprocessing object information to identify at least one object region and calculating an adapted detection threshold value. The adapted detection threshold value is equal to the default threshold value outside the at least one object region and equal to the at least one adapted threshold value within the at least one object region. The processor then uses the adapted detection threshold value to identify confirmation and candidate peaks in the range-Doppler-antenna data cube.

[0005] In some embodiments, the machine instructions for preprocessing the object information include machine instructions for converting the object position into radar coordinates for each tracked object in the object information and identifying the index of the range-Doppler segment corresponding to the radar coordinates. The processor then determines the area around the identified index to obtain at least one object region. In some embodiments, the machine instructions for calculating the adapted detection threshold include machine instructions for determining the object responsibility for each range-Doppler segment in the at least one object region and adjusting the default threshold based on the corresponding object responsibility to obtain the adapted threshold.

[0006] In some embodiments, the machine instructions for filtering the candidate peaks include machine instructions for determining the signal strength of the candidate peak for each candidate peak and determining whether the signal strength satisfies a first criterion. In response to the signal strength not satisfying the first criterion, the processor discards the candidate peak. In response to the signal strength satisfying the first criterion, the processor includes the candidate peak in the subset. The machine instructions for filtering the candidate peaks further include machine instructions for determining a similarity measure between the characteristics of the candidate peak and the corresponding characteristics of the corresponding tracked object in the object information for each candidate peak in the subset and determining whether the similarity measure satisfies a second criterion. In response to the similarity measure not satisfying the second criterion, the processor discards the candidate peak. In response to the similarity measure satisfying the second criterion, the processor includes the candidate peak in the point cloud. In some embodiments, the characteristics of the candidate peak and the corresponding characteristics of the corresponding tracked object in the object information include at least one of a distance, a speed, an azimuth, and an elevation angle.

[0007] A non-transitory computer-readable medium stores machine instructions that, when executed by at least one processor, cause the processor to perform a first FFT on received radar data to obtain a range-antenna data array, and to perform a second FFT on the range-antenna data array to obtain a range-Doppler-antenna data cube. The machine instructions cause the processor to perform peak detection on the range-Doppler-antenna data cube based on object information to obtain a subset of the range-Doppler-antenna data cube. The subset of the range-Doppler-antenna data cube includes confirmed peaks and candidate peaks. The machine instructions cause the processor to perform an angle of arrival calculation for the subset of the range-Doppler-antenna data cube, and to filter the candidate peaks in the subset of the range-Doppler-antenna data cube to obtain a point cloud representing an environment.

[0008] In some embodiments, confirmation peaks are detected based on a default threshold, and candidate peaks are detected based on at least one adapted threshold. At least one adapted threshold is determined based on object information. In some embodiments, machine instructions for performing peak detection on a range-Doppler-antenna data cube include machine instructions for preprocessing object information to identify at least one object region and calculating an adapted detection threshold. The adapted detection threshold is equal to the default threshold outside the at least one object region and equal to the at least one adapted threshold within the at least one object region. The machine instructions then cause the processor to use the adapted detection threshold to identify confirmation and candidate peaks in the range-Doppler-antenna data cube.

[0009] In some embodiments, the machine instructions for preprocessing the object information include machine instructions for converting the object position into radar coordinates for each tracked object in the object information and identifying the index of the range-Doppler segment corresponding to the radar coordinates. The machine instructions then cause the processor to determine a region around the identified index to obtain at least one object region. In some embodiments, the machine instructions for calculating the adapted detection threshold include machine instructions for determining an object responsibility for each range-Doppler segment in the at least one object region and adjusting a default threshold based on the corresponding object responsibility to obtain an adapted threshold.

[0010] In some embodiments, the machine instructions for filtering the candidate peaks include machine instructions for determining the signal strength of the candidate peak for each candidate peak and determining whether the signal strength satisfies a first criterion. In response to the signal strength not satisfying the first criterion, the machine instructions cause the processor to discard the candidate peak. In response to the signal strength satisfying the first criterion, the machine instructions cause the processor to include the candidate peak in the subset. The machine instructions for filtering the candidate peaks further include machine instructions for determining a similarity measure between the characteristics of the candidate peak and the corresponding characteristics of the corresponding tracked object in the object information for each candidate peak in the subset and determining whether the similarity measure satisfies a second criterion. In response to the similarity measure not satisfying the second criterion, the machine instructions cause the processor to discard the candidate peak. In response to the similarity measure satisfying the second criterion, the machine instructions cause the processor to include the candidate peak in the point cloud. In some embodiments, the characteristics of the candidate peak and the corresponding characteristics of the corresponding tracked object in the object information include at least one of distance, speed, azimuth, and elevation.

[0011] A method includes performing a first FFT on received radar data to obtain a range-antenna data array, and performing a second FFT on the range-antenna data array to obtain a range-Doppler-antenna data cube. Peak detection is performed on the range-Doppler-antenna data cube based on object information to obtain a subset of the range-Doppler-antenna data cube. The subset of the range-Doppler-antenna data cube includes confirmed peaks and candidate peaks. An angle of arrival calculation is performed for the subset of the range-Doppler-antenna data cube, and the candidate peaks in the subset of the range-Doppler-antenna data cube are filtered to obtain a point cloud representing an environment.

[0012] In some embodiments, confirmed peaks are detected based on a default threshold value, and candidate peaks are detected based on at least one adapted threshold value. At least one adapted threshold value is determined based on the object information. In some embodiments, performing peak detection on the range-Doppler-antenna data cube includes preprocessing the object information to identify at least one object region and calculating an adapted detection threshold value. The adapted detection threshold value is equal to the default threshold value outside the at least one object region and equal to the at least one adapted threshold value within the at least one object region. The adapted detection threshold value is used to identify confirmed and candidate peaks in the range-Doppler-antenna data cube.

[0013] In some embodiments, pre-processing the object information includes converting the object position into radar coordinates for each tracked object in the object information, and identifying, for each tracked object in the object information, an index of a range-Doppler segment corresponding to the radar coordinates. Determining a region around the identified index to obtain at least one object region. In some embodiments, calculating the adapted detection threshold includes determining an object responsibility for each range-Doppler segment in the at least one object region, and adjusting a default threshold based on the corresponding object responsibility for each range-Doppler segment in the at least one object region to obtain an adapted threshold.

[0014] In some embodiments, filtering the candidate peaks includes determining, for each candidate peak, a signal strength of the candidate peak and whether the signal strength satisfies a first criterion. In response to the signal strength not satisfying the first criterion, the candidate peak is discarded. In response to the signal strength satisfying the first criterion, the candidate peak is included in the subset. Filtering the candidate peaks further includes determining, for each candidate peak in the subset, a similarity measure between a characteristic of the candidate peak and a corresponding characteristic of a corresponding tracked object in the object information, and determining whether the similarity measure satisfies a second criterion. In response to the similarity measure not satisfying the second criterion, the candidate peak is discarded. In response to the similarity measure satisfying the second criterion, the candidate peak is included in the point cloud. In some embodiments, the characteristic of the candidate peak and the corresponding characteristic of the corresponding tracked object in the object information include at least one of a distance, a velocity, an azimuth, and an elevation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention may be better understood, and its numerous objects, features, and advantages made apparent to those skilled in the art by referencing the accompanying drawings.

[0016] Figure 1 An example radar system according to one embodiment is shown in block diagram form.

[0017] Figure 2 A diagram showing an example point cloud generated using conventional techniques and the disclosed technique for adjusting a detection threshold according to one embodiment.

[0018] Figure 3 A diagram illustrating an example data processing flow for generating a point cloud based on object information according to one embodiment is shown.

[0019] Figure 4 An example process for generating a point cloud based on object information according to one embodiment is shown in flowchart form.

[0020] Figure 5 An example process for performing peak detection using object information according to one embodiment is shown in flow chart form.

[0021] Figure 6 Graph showing object and noise probabilities, object responsibility, and adapted detection thresholds according to one embodiment.

[0022] Figure 7 An example process for filtering candidate peaks according to one embodiment is shown in flow chart form.

[0023] Unless otherwise indicated, the same reference numerals are used in different drawings to indicate the same items. The drawings are not necessarily drawn to scale. DETAILED DESCRIPTION

[0024] As previously discussed herein, some techniques for performing peak detection may result in missed detections where the signal strength of radar reflections from an object is close to the noise floor of the radar system. The disclosed techniques and radar systems implementing the disclosed techniques are capable of adapting the detection threshold based on information about the tracked object to selectively reduce the detection threshold in the zone where the object is expected to be located. Peaks detected by the reduced detection threshold are further processed to reduce the probability of missed detection or the false negative rate. Peaks detected by the reduced detection threshold can increase the detection density on the object. In addition, peaks detected by the reduced detection threshold are already associated with the tracked object and can better represent the extended object.

[0025] Figure 1 An example radar system 100 according to an embodiment is shown in block diagram form. The radar system 100 may be included in a vehicle, for example, for an automotive driver assistance system. The example radar system 100 is a frequency modulated continuous wave (FMCW) radar system, also known as a continuous wave frequency modulated (CWFM) radar, and is capable of determining the spacing or distance, velocity, and angle of arrival of objects in the field of view of the radar system 100. The term "angle of arrival" of an object is used herein to indicate the angle of arrival of a signal reflected from an object relative to the alignment of the radar system, and generally includes an azimuth and an elevation angle. Although the radar system 100 is described herein as a linear chirp radar system, any suitable radar system that generates range or Doppler information may be used.

[0026] In this example, the radar system 100 includes a microcontroller and processor unit (MCPU) 110, a digital-to-analog converter (DAC) 120, a radar sensor circuit 130, a first antenna array 145, a second antenna array 150, an analog-to-digital converter (ADC) 170, and a storage device 180. The MCPU 110 includes one or more MCPU cores, general processing cores, array or vector processing cores, parallel processing cores, graphics processing units, neural network and linear algebra accelerators, field programmable gate arrays, digital signal processors, application specific integrated circuits, etc., or any combination thereof. The singular term "MCPU" is used herein to refer to a single or multiple MCPU cores, general processing cores, array or vector processing cores, parallel processing cores, graphics processing units, digital signal processors, neural network and linear algebra accelerators, application specific integrated circuits, field programmable gate arrays, etc., or any combination thereof included in the MCPU 110.

[0027] The MCPU 110 includes a radar controller 115, a signal processor 175, and an object tracker 180. The radar controller 115 may receive data from the radar sensor circuit 130 and control radar parameters of the radar sensor circuit 130, such as the frequency band, length, etc. of the radar frame, via the DAC 120. The control signal from the DAC 120 may be used to adjust the radar chirp signal output from the chirp generator 135 included in the radar sensor circuit 130. The signal processor 175 in the MCPU 110 may also receive data from the radar sensor circuit 130 and perform signal processing for determining a spacing or distance between a target object and the radar system 100, a radial velocity of the target object, an angle of arrival of the target object, etc. The signal processor 175 may provide the calculated values ​​to the storage device 185 and / or other systems via the interface 190.

[0028] The object tracker 180 may obtain object information from the storage device 185 or from other types of sensing systems via the interface 190. For example, the object tracker 180 may obtain object information from an imaging or stereoscopic imaging system, a light detection and ranging (LiDAR) system, etc., and a combination of sensing systems. In this embodiment, the object tracker 180 is included in the MCPU 110, but in other embodiments, the object tracker 180 may be included in a separate processing system and provide object information to the signal processor 175 via the interface 190. The object information may include the number of objects in the environment (if any); and the dynamic state of the objects in the environment, including at least one of the position, velocity, acceleration, turn rate, etc. of the objects in the environment in polar coordinates, Cartesian coordinates, or other coordinates. The object information may include information based on sensor data captured at a previous time, such as a radar frame before the current radar frame. Additionally or alternatively, the object information may include information based on sensor data captured simultaneously with the current radar frame. For example, different sensing systems such as camera systems and the radar system 100 capture data at approximately the same time. The object detection performed on the camera data is completed before the object detection performed on the radar data is completed, so that the object information includes camera data captured at approximately the same time as the radar data.

[0029] Interface 190 may enable MCPU 110 to communicate with other systems, for example, via local and wide area networks, the Internet, a vehicle communication bus, and / or other types of wired or wireless communication systems. MCPU 110 may provide calculated values ​​to other systems, such as radar-camera-lidar fusion systems; automated driving assistance systems including parking, braking, or lane change assistance features; and the like, through interface 190. Storage device 185 may be used to store instructions for MCPU 110, data received from radar sensor circuit 130, values ​​calculated from signal processor 175, and the like. Storage device 185 may be any suitable storage medium, such as volatile or non-volatile memory.

[0030] The radar sensor circuit 130 includes a chirp generator 135, a transmitter 140, a receiver 155, and a baseband processor 160. The chirp generator 135 may include, for example, a local oscillator, and generates a radar chirp signal and provides it to the transmitter 140. For example, the frequency of the chirp generator 135 may modulate a continuous wave signal to form a series of linear chirp signals. The frequency of the transmitted chirp signal of the known stable frequency continuous wave varies up and down within a certain period of time by the modulated signal. The chirp generator 135 provides the generated chirp signal to the transmitter 140, which drives the first antenna array 145 of one or more transmitter (TX) antennas. The second antenna array 150 includes one or more receiver (RX) antennas and receives signals reflected from objects in the path of the transmitted chirp signal from the TX antenna array 145. The TX antenna array 145 and the RX antenna array 150 may be stationary or configured to transmit and receive across a series of areas, for example, by mechanical movement.

[0031] The receiver 155 receives the reflected signal from the RX antenna array 150 and provides it to the baseband processor 160. The baseband processor 160 also receives the transmitted chirp signal from the chirp generator 135 and down-converts the received chirp signal directly into baseband using a copy of the transmitted chirp signal from the chirp generator 135. The baseband processor 160 may then filter and amplify the baseband signal. The baseband processor 160 provides the filtered and amplified baseband signal to the ADC 170, which digitizes the signal and provides it to the MCPU 110. The signal processor 175 in the MCPU 110 may then perform a time domain to frequency domain transform, such as a fast Fourier transform (FFT) and other signal processing to determine the spacing, radial velocity, and angle of arrival between the target object and the radar system 100.

[0032] The frequency difference between the received reflection and the transmitted chirp signal increases with increasing delay and is therefore proportional to the spacing. The phase difference between the received reflections of consecutive chirps across the radar frame indicates the speed of the object in the field of view. For embodiments in which the RX antenna array 150 includes two or more receiver antennas, the phase difference between the reflection received at the first RX antenna and the reflection received at the second RX antenna can be used to determine the angle of arrival of the target object. For example, the down-converted and digitized received signal corresponding to each chirp is first transformed using an FFT (referred to as a range FFT). The range FFT produces a series of range segments, where the absolute value of each range segment represents the signal strength of the reflected target at the corresponding distance. Then, another "Doppler" FFT is performed for each range segment across all chirps in the frame to estimate the speed of the reflected target. Then, additional processing can be performed to determine the angle of arrival between the target and the radar system 100.

[0033] Although radar system 100 is described herein as implementing FFT-based range-Doppler processing, any suitable transform may be used to generate range-Doppler information. Radar system 100 may then output a point cloud representing the environment surrounding radar system 100, and each point in the point cloud may include information indicating the distance between radar system 100 and the point in the environment, the radial velocity of the point, the azimuth of the point relative to radar system 100, the elevation of the point relative to radar system 100, and any combination thereof.

[0034] For embodiments in which the radar system 100 is included in a vehicle having an automated driving assistance system, the automated driving assistance system may use the distance, speed, and angle of arrival of objects in the field of view determined from the radar system 100 to provide parking, braking, or lane change assistance. The radar system 100 must detect objects in the field of view of the radar system 100 and determine the corresponding distance, speed, and / or angle of arrival. The radar system according to the disclosed invention adjusts the object detection threshold based on the object information.

[0035] Figure 2A diagram 200 of an example point cloud generated using conventional techniques and a diagram 250 of an example point cloud generated using the disclosed technique for adjusted detection thresholds are shown according to one embodiment. Diagram 200 includes points generated using conventional techniques and a constant detection threshold within each distance segment without reference to object information indicated by zones 210, 220, and 230. For example, zones 210, 220, and 230 may indicate areas in the point cloud where an object was detected at an earlier time, such as a radar frame prior to the current radar frame. As another example, different sensing systems, such as a camera system, and a radar system capture data at approximately the same time. Object detection performed on camera data is completed before object detection performed on radar data is completed, so that zones 210, 220, and 230 may be object information determined from camera data captured at approximately the same time as radar data. In contrast, diagram 250 includes points generated using detection thresholds adjusted based on object information indicated by zones 260, 270, and 280. For example, for areas of the point cloud outside of zones 260, 270, and 280, the adjusted detection threshold may be equal to the constant detection threshold within each distance segment used to generate graph 200, and within zones 260, 270, and 280, the adjusted detection threshold may be reduced based on the probability of an object at a particular point. Due to the lower adjusted detection threshold in zones 260, 270, and 280, multiple additional points are detected—two additional points are detected in zone 260, three additional points are detected in zone 270, and six additional points are detected in zone 280.

[0036] The adjusted detection threshold used to generate diagram 250 can be used to better identify extended objects than the fixed detection threshold of diagram 200. That is, the adjusted detection threshold can be used to more effectively identify objects that occupy more than a single distance or speed segment than the fixed detection threshold. For the example where radar system 100 is included on a vehicle as part of an automated driving assistance system, the adjusted detection threshold can be used to identify long vehicles that occupy multiple distance segments, such as semi-trucks. Extended objects produce signals from radar reflections that span multiple segments, which conventional techniques can interpret as increased background noise and therefore increase the detection threshold. In contrast, the disclosed techniques for adjusted detection thresholds take full advantage of the object information indicated by zones 260, 270, and 280 to more accurately resolve object detections.

[0037] Figure 3 Graphs 300 and 350 are shown illustrating an example data processing flow for generating a point cloud based on previously tracked object information according to one embodiment. Figure 1300 and 350 are described with reference to the radar system 100 shown in FIG. 300 shows a signal processing flow for a digitized received radar signal 310A at a first time t0. The signal processor 175 performs a range FFT 315 on the digitized received radar signal 310A, and then the signal processor 175 performs a Doppler FFT 320 on the output of the range FFT 315. The signal processor 175 then performs a peak detection 330 based on object information 325. The object information 325 may include the number of objects in the environment (if any) and the dynamic state information of the objects, as previously described herein with reference to FIG. Figure 1 Object information 325 may include information indicating at least the position and velocity of a previously identified object in the environment. For example, object information 325 may include the position (p) of a previously identified object in the environment in Cartesian coordinates. x ,p y ) and speed (v x ,v y In some embodiments, object information 325 may include additional information about previously identified objects, such as height p z The object information may come from the radar system 100 , an imaging system, a lidar system, etc., or any combination thereof.

[0038] To perform peak detection 330, signal processor 175 may use object information 325 to determine an adapted threshold value that is equal to a default threshold value for a range-Doppler segment outside of a tracked object zone indicated by object information 325, and that is equal to a lower adapted threshold value inside a tracked object zone indicated by object information 325. After performing peak detection 330, signal processor 175 may perform an angle of arrival calculation 335 to determine an azimuth and / or elevation angle of an object detected in the environment. The output of angle of arrival calculation 335 is a point cloud 345A that may include information about the range, Doppler, azimuth, and elevation angle of the detection point associated with time t0. Point cloud 345A may be used to generate updated object information 340. That is, signal processor 175 or object tracker updates object information 325 to include object information associated with radar data at time t0.

[0039] Graph 350 is substantially similar to graph 300 and illustrates a signal processing flow for a digitized received radar signal 310B at a second subsequent time t1. Signal processor 175 performs range FFT 315 and Doppler FFT 320 and then performs peak detection 330 based on updated object information 340 including object information associated with time t0. To perform peak detection 330, signal processor 175 may use object information 340 to determine an adapted threshold value that is equal to a default threshold value for a range-Doppler segment outside of a tracked object region indicated by object information 340 and that is equal to a lower adapted threshold value inside a tracked object region indicated by object information 340.

[0040] After performing peak detection 330, signal processor 175 may perform an angle of arrival calculation 335 to determine the azimuth and / or elevation of an object detected in the environment. The output of angle of arrival calculation 335 is a point cloud 345B that may include information about the distance, Doppler, azimuth, and elevation of the detection point associated with time t1. Point cloud 345B may be used to generate updated object information 355. That is, signal processor 175 or object tracker updates object information 340 to include object information associated with radar data at time t1.

[0041] Figure 4 An example process 400 for generating a point cloud based on object information according to one embodiment is shown in flowchart form. Figure 1 The radar system 100 and Figure 2 The diagram 250 shown in FIG. 4 depicts a process 400. The process 400 is described herein as being performed by the signal processor 175 in the MCPU 110 executing instructions stored in the memory device 185 of the radar system 100, but may be performed by any suitable processing unit executing instructions stored in any suitable non-transitory computer-readable medium. The steps of the process 400 are described in detail in FIG. Figure 4 Although shown in a particular order in the process 400, the steps of process 400 may be performed in a different order and / or some steps may be performed simultaneously. One or more steps of process 400 may be optional, and process 400 may include additional steps.

[0042] The process begins at step 410, where the signal processor 175 performs a range FFT on the digitized received radar data to obtain a range-antenna data array 415. At step 420, the signal processor 175 performs a Doppler FFT on the range-antenna data array 415 to obtain a range-Doppler-antenna data cube 425. At step 430, the signal processor 175 performs peak detection on the range-Doppler-antenna data cube 425 using the object information. For example, the signal processor 175 may obtain the object information from an object tracker 180 included in the radar system 100, from a storage device 185, or from an object tracker or storage device not included in the radar system 100 via the interface 190. Figure 5 The output of the peak detection is a subset 435 of the range-Doppler-antenna data cube 425, including "confirmed peaks" - peaks detected outside the tracked object regions 260, 270, and 280 shown in the diagram 250 using the default threshold, and "candidate peaks" - peaks detected inside the tracked object regions 260, 270, and 280 using a lower adapted threshold.

[0043] The signal processor 175 performs an angle of arrival calculation on the subset 435 of the range-Doppler-antenna data cube 425 at step 440 to determine the azimuth and / or elevation angles of the confirmed peaks and candidate peaks in the subset 435. At step 450, the signal processor 175 filters (filters) the candidate peaks in the subset 435 of the range-Doppler-antenna data cube 425. In some embodiments, the signal processor 175 filters (filters) the candidate peaks based on corresponding signal-to-noise ratio (SNR), range, Doppler, azimuth, elevation, or a combination thereof. Although in process 400, the filtering step 450 is shown as being performed after the angle of arrival calculation step 440, the filtering (filtering) step 450 may be performed at other locations in process 400.

[0044] For example, the candidate peaks are filtered based on the SNR and the azimuth. The signal processor 175 may perform a first filtering step 450A after the peak detection step 430 to filter the candidate peaks based on the SNR. The angle of arrival calculation step 440 may then be performed for fewer candidate peaks, thereby reducing the computational cost of step 440. A second filtering step 450B may be performed after the angle of arrival calculation step 440 to filter the candidate peaks based on the determined azimuth. The result of the filtering step 450B after the angle of arrival calculation step 440 is a point cloud 455, which may be provided to, for example, an automated driving assistance system. The signal processor 175 then updates the object information based on the point cloud 455 in step 460.

[0045] Figure 5An example process 500 for performing peak detection using object information according to one embodiment is shown in flowchart form. For example, process 500 may be used to implement Figure 4 The peak detection step 430 of the process 400 shown in FIG. Figure 1 The radar system 100 and Figure 6 Schematics 600, 630, and 650 shown in the figure describe process 500. Process 500 is described herein as being performed by signal processor 175 in MCPU 110 executing instructions stored in memory device 185 of radar system 100, but may be performed by any suitable processing unit executing instructions stored in any suitable non-transitory computer-readable medium. The steps of process 500 are described in detail in FIG. Figure 5 Although shown in a particular order in the process 500, the steps of process 500 may be performed in a different order and / or some steps may be performed simultaneously. One or more steps of process 500 may be optional, and process 500 may include additional steps.

[0046] The process begins at step 510, where the signal processor 175 pre-processes the object information. Step 510 includes optional step 520 and steps 530 and 540. In optional step 520, the signal processor 175 converts the object information into radar coordinates representing range R, Doppler D, and azimuth θ. For example, the object information is presented in Cartesian coordinates and includes position and velocity information of each object included in the object information. That is, the object information includes state information x of each object. cart . State information of a specific object in Cartesian coordinates x cart It can be expressed as:

[0047]

[0048] Among them (p x ,p y ) represents the location of a particular object in Cartesian coordinates, and (v x ,v y ) represents the velocity of a particular object in Cartesian coordinates.

[0049] Due to the probabilistic nature of estimation theory, the state information x of a particular object cart It is represented by the probability distribution p(x) in the object information. Some object information may use Gaussian distribution, so that the probability distribution p(x) can be expressed as:

[0050]

[0051] in Mean parameter The Gaussian probability density function and the covariance matrix P cartExpressed as E[·] denotes the expectation operator. Although Gaussian distribution is used herein, any suitable probability distribution may be used.

[0052] Status information of a specific object in radar coordinates x radar It can be expressed as:

[0053]

[0054] The nonlinear mapping between Cartesian coordinates and radar coordinates is represented by x radar =h(x cart ).

[0055] In order to convert the state estimate p(x) of a specific object from the object information in Cartesian coordinates into cart ) is converted into the state estimate p(x) in radar coordinates (R, D, θ) radar ), the signal processor 175 performs the following transformation:

[0056] μ radar =h(μ cart )

[0057] P radar ≈HP radar H T +N,

[0058]

[0059] Where H is the nonlinear transformation h(μ cart ), N is the sensor covariance matrix representing the measurement noise of the radar system 100, represents the noise variance of the distance measurement, represents the noise variance of the Doppler measurement, and represents the noise variance of the azimuth measurement. Therefore, the state estimate p(x) of the object in Cartesian coordinates is cart ) is determined by the uncertainty around the point μ cart The linear approximation h(·) propagates to the state estimate p(x radar ), and taking into account measurement noise. In this example, the mean and uncertainty refer to the center of the tracked object. In some embodiments, the range of the object is also estimated so that the uncertainty of the state estimate p(x) also captures the range of the object. That is, the uncertainty of the state estimate p(x) can also represent the uncertainty about the number of range segments occupied by the object. For object information that is already in radar coordinates, optional step 520 can be omitted.

[0060] At step 530, the signal processor 175 identifies the index (Ir, Id) of the range-Doppler segment corresponding to the radar coordinates (R, D, θ). That is, for each object included in the object information, the signal processor 175 identifies the range segment index Ir and the Doppler segment index Id corresponding to the determined radar coordinates of the object. At step 540, the signal processor 175 identifies the area around the index (Ir, Id) to determine the object area. For example, the signal processor 175 determines the areas 260, 270, and 280 in the diagram 250. In some embodiments, the covariance matrix P may be used to determine the object area. radar And select the number of distance segments n around the index (Ir, Id) r and the number of Doppler segments n d For example, by replacing the covariance matrix P radar The square root of the first diagonal element of is divided by the range resolution of radar system 100 and the covariance matrix P radar The number of range segments n is determined by dividing the square root of the second diagonal element of by the Doppler resolution of the radar system 100. r and the number of Doppler segments n d In some embodiments, the number of distance segments n r It can be a fixed value to reduce the computational complexity. For example, the number of distance segments n r can be set equal to five, so that the tracked object area includes the tracked object area having index [i r -2,i r -1,i r ,i r +1,i r +2] distance segment.

[0061] At step 550, the signal processor 175 calculates the adapted detection threshold. Step 550 includes steps 560 and 570. At step 560, the signal processor 175 determines the object probability density for each range-Doppler segment in the object region. The object probability density in a particular range gate in the object region can be approximated as where μ d μ radar The second element of P radar The signal processor 175 can then use the object probability density p object (d) The segmentation probability Pr of the object object [j] is calculated as:

[0062]

[0063] where j represents Doppler segment j, and i,land Β i,u Respectively represent the lower and upper values ​​of the jth Doppler segment. objeect [j] is the probability that the center of the object is in a particular Doppler segment j and may be calculated using a Gaussian distribution or other numerically approximated cumulative distribution function.

[0064] In order to take into account the noise in the radar signal, a uniform noise distribution is used so that the noise segment probability Pr noise [j] is constant within the Doppler segment range. That is, the noise segment probability Pr noise [j] can be expressed as:

[0065]

[0066] Therefore, the total probability distribution of reflection Pr total [j] is represented by:

[0067]

[0068] where w noise is the weight associated with the noise distribution, the number of objects M is included in the object information, and is the weight associated with the object distribution of a particular object m. noise and are non-negative and sum to 1. The weight w noise and The specific value of may be selected based on the specific implementation. noise as well as The sum can be set equal to a fixed value, The fixed value of the sum can be the weight of the M objects in the object information is equally divided between so that the probability of each object is the same. In other embodiments, The fixed value of the sum can be the weight of the M objects in the object information The weight w is divided between the two to reflect the relative probability of existence of the objects, with objects with higher confidence of existence based on previous measurements being given greater weight than objects with lower confidence of existence. noise and w object It can be adaptively calculated based on, for example, the existence probability of a specific object, the SNR of a specific object, etc. For example, an object with a probability greater than 50% can make w object Greater than w noise , and there are objects with probability less than 50% that can make w object Less than w noise .

[0069] To show the relative weight of noise reflections and radar reflections in the signal, Figure 6The diagram 600 shown in FIG. 6 represents the object segmentation probability Pr object [j]610 and noise segment probability Pr noise [j] 620 versus velocity. In diagram 600, radar system 100 implements a Doppler-division multiplexing scheme, resulting in a single object being represented by two velocities - in this example, approximately -4 meters per second (m / s) and approximately 11 m / s, which are disambiguated by a Doppler disambiguation process. A negative velocity indicates that the object is moving toward radar system 100, while a positive velocity indicates that the object is moving away from radar system 100. Noise segmentation probability Pr noise [j] 620 is constant over the speed range to reflect the uniform noise distribution. object [j] 610 includes peaks at approximately -4 m / s and at approximately 11 m / s, indicating that at those speeds, radar reflections from objects account for a much larger portion of the radar signal than noise.

[0070] Object responsibility R per segment [object] [j] is the object segmentation probability Pr object [j] accounts for the total probability distribution Pr total The weight of [j] divided by the ratio of the probability of the combination. That is, the object responsibility R of each segment [object] [j] can be expressed as:

[0071]

[0072] Object responsibility R per segment [object] Each value of [j] shows the contribution of radar reflections from tracked objects to the total measured radar signal in a particular Doppler segment j. [object] [j] can be a number between 0 and 1, so that multiplying by 100 gives the percentage contribution of the radar reflection from the tracked object to the total measured radar signal in a particular Doppler segment j. Figure 6 The graph 630 shown in FIG. 600 shows the object segmentation probability Pr object [j]610 and noise segment probability Pr noise [j] 620 Object responsibility R per segment [object] [j] 640. That is, the object segmentation probability Pr object [j] 610 relative to the noise segment probability Pr noise [j] The proportion of 620 in the total probability distribution Pr total The specific gravity of [j] is much higher at a speed of about -4 m / s and at about 11 m / s than at other speeds.

[0073] At step 570, the signal processor 175 adjusts the default detection threshold for each range-Doppler segment in the tracked object region based on the corresponding object responsibility. default The detection threshold Thr may be lowered by an amount based on the contribution of the object to that particular Doppler segment. adapted [j] can be expressed as:

[0074] Thr adapted [j]=Thr default [j]-α*R [object] [j]

[0075] Where α represents the maximum permissible reduction in the detection threshold. The maximum permissible reduction α may be selected based on the SNR characteristics of radar system 100, a dedicated balance between false alarms and detection probability, etc. Figure 6 The diagram 650 shown in FIG. 6 shows a radar signal 660, a default detection threshold Thr default 670 and the adapted detection threshold Thr adapted 680. At speeds outside the object area, the default detection threshold Thr default 670 and the adapted detection threshold Thr adapted 680. At the speed in the object area, based on the object responsibility R per segment shown in the diagram 630 [object] [j] 640, the detection threshold Thr has been adjusted adapted 680 has a higher detection threshold than the default Thr default 670 low value. For example, at a speed of about -5m / s, the detection threshold Thr has been adapted adapted The value of 680 is slightly lower than the default detection threshold Thr default 670, and at a speed of about -4m / s, the detection threshold Thr has been adjusted adapted The value of 680 is higher than the default detection threshold Thr default 670 is much lower. Compared with -5m / s, the default detection threshold Thr at -4m / s default 670 and the adapted detection threshold Thr adapted The greater difference between the values ​​of 680 reflects the object responsibility R per segment at -4 m / s compared to -5 m / s [object] [j]640 is bigger.

[0076] At step 580, the signal processor 175 uses the adapted detection threshold calculated at step 550 to identify confirmed peaks and candidate peaks in the range-Doppler-antenna data cube 425. For ease of explanation, a confirmed peak is a peak detected outside the object region where the adapted detection threshold has a value equal to the default detection threshold. A candidate peak is a peak detected within the object region where the adapted detection threshold has a value less than the default detection threshold. For example, at Figure 2 In the diagram 250 shown in FIG. 4 , points detected using an adjusted detection threshold are considered candidate peaks, while points detected using a constant detection threshold are considered confirmed peaks. Candidate peaks may be filtered at step 450 of process 400 as previously described herein to prevent false positives from being introduced into point cloud 455. At step 590, signal processor 175 includes the confirmed peaks and candidate peaks in subset 435 of range-Doppler-antenna data cube 425.

[0077] Figure 7 An example process 700 for filtering candidate peaks according to one embodiment is shown in flowchart form. For example, process 700 may be used to implement Figure 4 The candidate peak filtering step 450 of the process 400 shown in FIG. Figure 1 The process 700 is described with reference to the radar system 100 shown in FIG. The process 700 is described herein as being performed by the signal processor 175 in the MCPU 110 executing instructions stored in the memory device 185 of the radar system 100, but may be performed by any suitable processing unit executing instructions stored in any suitable non-transitory computer-readable medium. The steps of the process 700 are described in FIG. Figure 7 Although shown in a particular order in the process 700, the steps of process 700 may be performed in a different order and / or some steps may be performed simultaneously. One or more steps of process 700 may be optional, and process 700 may include additional steps.

[0078] The process begins at step 705, where the signal processor 175 determines a signal strength for a candidate peak. At step 710, the signal processor 175 determines whether the determined signal strength satisfies an SNR criterion. The SNR criterion may be a minimum SNR, such that the signal processor 175 determines whether the determined signal strength is greater than the minimum SNR. In response to determining that the signal strength does not satisfy the SNR criterion, the signal processor 175 discards the candidate peak at step 715. That is, the candidate peak is not sufficiently different from the noise in the radar signal and would introduce confusion if included in the point cloud 455. In response to determining that the signal strength satisfies the SNR criterion, the signal processor 175 includes the candidate peak in the subset at step 720. The signal processor 175 then determines at step 725 whether there are any remaining candidate peaks. In response to determining that there are remaining candidate peaks, the signal processor 175 returns to step 705 to analyze the next candidate peak.

[0079] In response to determining that there are no remaining candidate peaks, the signal processor 175 proceeds to step 730 and determines a similarity measure between the candidate peak and the characteristic of the corresponding tracked object in the previous tracked object information. For example, the signal processor 175 may determine the L1 norm between the value of the characteristic of the candidate peak and the value of the characteristic of the corresponding tracked object. The characteristics of the candidate peak and the corresponding tracked object may be a Doppler value 735, an azimuth angle 740, an elevation angle 745, and a distance 750. In some embodiments, the signal processor 175 determines a similarity measure between more than one characteristic. For example, the signal processor 175 determines a first similarity measure for the Doppler value 735 and a second similarity measure for the azimuth angle 740. In some embodiments, the radar system 100 has low accuracy and / or resolution for the elevation angle 745, so that the elevation angle 745 is not a useful characteristic for filtering the candidate peak, and the signal processor 175 does not consider the elevation angle 745. The distance 750 is already similar between the candidate peak and the corresponding tracked object because the candidate peak is determined based on a reduced detection threshold in the tracked object area. In some embodiments, signal processor 175 does not take distance 750 into account to reduce computational complexity.

[0080] At step 755, the signal processor 175 determines whether one or more similarity measures satisfy a criterion, such as a maximum difference between the candidate peak and the characteristic of the corresponding tracked object. In response to the similarity measure not satisfying the criterion, such as in response to the candidate peak and the characteristic of the corresponding tracked object exceeding the maximum difference, the signal processor 175 discards the candidate peak at step 760. In response to the similarity measure satisfying the criterion, such as in response to the candidate peak and the characteristic of the corresponding tracked object being within the maximum difference, the signal processor 175 includes the candidate peak in the point cloud 455 at step 765. The signal processor 175 then determines at step 770 whether there are any remaining candidate peaks in the subset. In response to determining that there are remaining candidate peaks, the signal processor 175 returns to step 730 to analyze the next candidate peak. In response to determining that there are no remaining candidate peaks in the subset, the signal processor 175 outputs the point cloud 455.

[0081] As previously mentioned in this article Figure 4 As described in step 450 of process 400 shown in , the steps of filtering process 700 can be interspersed among the steps of process 400 to reduce the number of candidate peaks to complete the steps of process 400. For example, steps 705-725 can be performed immediately after the candidate peaks are identified in step 430 to filter out the candidate peaks based on the corresponding signal strength. Similarly, steps 730-770 can be performed for distance 750 immediately after the candidate peaks are identified in step 430 to filter out the candidate peaks based on the corresponding distance. Therefore, the number of candidate peaks for disambiguating Doppler values ​​and calculating the angle of arrival is reduced. Steps 730-770 can be performed for the disambiguated Doppler value 735 after performing the Doppler disambiguation process, thereby reducing the number of candidate peaks for calculating the angle of arrival.

[0082] As described herein, the disclosed technique for adapting a peak detection threshold based on previously tracked object information reduces the number of missed detections by lowering the detection threshold in a region around a tracked object included in the previously tracked object information, thereby locally increasing detection sensitivity. Strategic filtering of candidate peaks reduces the number of false positives and clutter introduced into the resulting point cloud by a lower detection threshold. Strategic filtering also reduces the number of candidate peaks for which additional calculations are performed, thereby reducing the computational cost of additional peaks identified by a lower detection threshold. Additionally, the adapted detection threshold enables the resulting point cloud to better represent extended objects that occupy more than a single distance segment.

[0083] Features specifically shown or described with respect to one embodiment described herein may be implemented on other embodiments described herein.

[0084] In this specification, the term "coupling" may encompass connections, connectivity or signal paths that implement functional relationships according to this specification, including intermediate or intervening components that do not change the functional relationship. A device "configured to" perform a task or function may be configured, for example, by programming or hardwiring when the manufacturer manufactures it, and / or the device may be configured or reconfigured by the user after manufacture. The configuration may be accomplished by firmware and / or software, hardware components and the construction and / or layout of the connections, or any combination thereof. As used herein, "node", "pin" and "lead" are used interchangeably. A circuit or device described herein as including certain components may be adapted to be alternatively coupled to those components to form the described circuit system or device.

[0085] While particular embodiments of the present invention have been shown and described, those skilled in the art will recognize that, based on the teachings herein, further changes and modifications may be made without departing from the present invention and its broader aspects, and therefore, the appended claims are intended to cover within their scope all such changes and modifications that are within the true spirit and scope of the present invention.

Claims

1. A system, characterized in that: include: at least one processor; as well as at least one non-transitory computer-readable medium storing machine instructions that, when executed by the at least one processor, cause the at least one processor to: Performing a first fast Fourier transform (FFT) on the received radar data to obtain a range-antenna data array; performing a second FFT on the range-antenna data array to obtain a range-Doppler-antenna data cube; performing peak detection on the range-Doppler-antenna data cube based on object information to obtain a subset of the range-Doppler-antenna data cube, wherein the subset of the range-Doppler-antenna data cube includes confirmed peaks and candidate peaks; performing an angle of arrival calculation for said subset of said range-Doppler-antenna data cube; as well as Candidate peaks in the subset of the range-Doppler-antenna data cube are filtered to obtain a point cloud representing an environment.

2. The system according to claim 1, characterized in that: detecting the confirmation peak based on a default threshold; detecting the candidate peak based on at least one adapted threshold; and The at least one adapted threshold is determined based on the object information.

3. The system according to claim 2, characterized in that The machine instructions to perform peak detection on the range-Doppler-antenna data cube include machine instructions to: preprocessing the object information to identify at least one object region; calculating an adapted detection threshold, wherein the adapted detection threshold is equal to the default threshold outside the at least one object region and is equal to the at least one adapted threshold within the at least one object region; as well as The adapted detection threshold is used to identify candidate peaks in the range-Doppler-antenna data cube.

4. The system according to claim 1, characterized in that The machine instructions for filtering candidate peaks include machine instructions for performing the following operations: For each candidate peak: Determining the signal strength of the candidate peak; Determining whether the signal strength meets a first criterion; In response to the signal strength not satisfying the first criterion, discarding the candidate peak; as well as In response to the signal strength satisfying the first criterion, including the candidate peak in a subset; For each candidate peak in the subset: determining a similarity measure between a characteristic of the candidate peak and a corresponding characteristic of a corresponding tracked object in the object information; determining whether the similarity measure satisfies a second criterion; In response to the similarity measure not satisfying the second criterion, discarding the candidate peak; as well as In response to the similarity measure satisfying the second criterion, the candidate peak is included in the point cloud.

5. A non-transitory computer-readable medium, characterized in that storing machine instructions that, when executed by at least one processor, cause the at least one processor to: Performing a first fast Fourier transform (FFT) on the received radar data to obtain a range-antenna data array; performing a second FFT on the range-antenna data array to obtain a range-Doppler-antenna data cube; performing peak detection on the range-Doppler-antenna data cube based on object information to obtain a subset of the range-Doppler-antenna data cube, wherein the subset of the range-Doppler-antenna data cube includes confirmed peaks and candidate peaks; performing an angle of arrival calculation for said subset of said range-Doppler-antenna data cube; as well as Candidate peaks in the subset of the range-Doppler-antenna data cube are filtered to obtain a point cloud representing an environment.

6. The non-transitory computer readable medium of claim 5, wherein: detecting the confirmation peak based on a default threshold; detecting the candidate peak based on at least one adapted threshold; and The at least one adapted threshold is determined based on the object information.

7. The non-transitory computer-readable medium of claim 5, wherein: The machine instructions for filtering candidate peaks include machine instructions for performing the following operations: For each candidate peak: Determining the signal strength of the candidate peak; Determining whether the signal strength meets a first criterion; In response to the signal strength not satisfying the first criterion, discarding the candidate peak; as well as In response to the signal strength satisfying the first criterion, including the candidate peak in a subset; For each candidate peak in the subset: determining a similarity measure between a characteristic of the candidate peak and a corresponding characteristic of a corresponding tracked object in the object information; determining whether the similarity measure satisfies a second criterion; In response to the similarity measure not satisfying the second criterion, discarding the candidate peak; as well as In response to the similarity measure satisfying the second criterion, the candidate peak is included in the point cloud.

8. A method, characterized in that include: Performing a first fast Fourier transform (FFT) on the received radar data to obtain a range-antenna data array; performing a second FFT on the range-antenna data array to obtain a range-Doppler-antenna data cube; performing peak detection on the range-Doppler-antenna data cube based on object information to obtain a subset of the range-Doppler-antenna data cube, wherein the subset of the range-Doppler-antenna data cube includes confirmed peaks and candidate peaks; performing an angle of arrival calculation for said subset of said range-Doppler-antenna data cube; as well as Candidate peaks in the subset of the range-Doppler-antenna data cube are filtered to obtain a point cloud representing an environment.

9. The method according to claim 8, characterized in that: detecting the confirmation peak based on a default threshold; detecting the candidate peak based on at least one adapted threshold; and The at least one adapted threshold is determined based on the object information.

10. The method according to claim 8, characterized in that Filtering of candidate peaks includes: For each candidate peak: Determining the signal strength of the candidate peak; Determining whether the signal strength meets a first criterion; In response to the signal strength not satisfying the first criterion, discarding the candidate peak; and In response to the signal strength satisfying the first criterion, including the candidate peak in a subset; For each candidate peak in the subset: determining a similarity measure between a characteristic of the candidate peak and a corresponding characteristic of a corresponding tracked object in the object information; determining whether the similarity measure satisfies a second criterion; In response to the similarity measure not satisfying the second criterion, discarding the candidate peak; and In response to the similarity measure satisfying the second criterion, the candidate peak is included in the point cloud.

Images