A method for improving the maximum perception speed of high-resolution millimeter wave radar

Abstract

The application discloses a method for improving the maximum sensing speed of high-resolution millimeter wave radar, which comprises the following steps: firstly, configuring parameters on a millimeter wave radar evaluation board and configuring a transmission mode; secondly, designing a signal processing system, which will process signals received by the millimeter wave radar frame by frame to obtain a point cloud dataset, and then respectively clustering the point cloud dataset to obtain a point cloud clustering dataset; then, according to the point cloud clustering dataset, a point cloud matching dataset containing matched and unmatched points is obtained by using BEV information, and an error matrix scheme is selected to calculate and obtain a defuzzification strategy with minimum error; finally, according to the defuzzification strategy with minimum error, the point cloud matching dataset is defuzzified, and the defuzzified radar point cloud result is drawn on an optical photo. The application solves the problem that the high-resolution millimeter wave radar for assisting intelligent driving is limited by spatial resolution and speed resolution, and realizes a wide sensing speed detection range.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of millimeter wave radar measurement, and particularly relates to a method for improving the maximum perception speed of a high-resolution millimeter wave radar. BACKGROUND

[0002] Millimeter wave radar has advantages that laser radar does not have under special weather conditions. According to wave propagation theory, the higher the frequency, the shorter the wavelength, the higher the resolution, and the stronger the penetration ability. Compared with other microwaves, millimeter waves have high resolution, good directivity, strong anti-interference ability and good detection performance, especially in the field of intelligent driving. In complex road conditions and bad weather, target recognition, analysis, perception and information transmission are more accurate and suitable compared with traditional laser radar. Under such technical background, the research on millimeter wave radar with high resolution and adaptability to high-speed scenes has great significance for industrial perception and application.

[0003] Although millimeter wave radar has the above-mentioned significant advantages, due to hardware limitations, it faces problems such as the inability of high-resolution performance and wide perception speed range to coexist. Therefore, the high spatial and speed resolution millimeter wave radar used today is limited to a narrow perception speed detection range, and a wide perception speed range will bring performance sacrifices such as spatial and speed resolution. In order to meet the high-precision, high-resolution and high-speed application requirements of automatic driving on millimeter wave radar, the demand for improving the perception speed range of high-resolution radar is put forward.

[0004] At present, millimeter wave radar still has the problem that high-resolution performance and wide perception speed range cannot coexist. If the maximum perception speed range can be improved from the signal processing level without sacrificing the spatial and speed resolution of the millimeter wave radar, it will directly provide a feasible solution for the application of millimeter wave radar into high-speed scenes such as highways. It will have important significance for all-weather auxiliary traffic on highways, ramp merging and splitting warning, fog area driving induction and other application scenarios. SUMMARY

[0005] The application provides a method for improving the maximum perception speed of a high-resolution millimeter wave radar, in order to solve the contradiction between the maximum measurement speed of spatial resolution and the contradiction between the maximum measurement speed of speed resolution.

[0006] The technical scheme adopted by the application is as follows:

[0007] Step 1, configure parameters on a millimeter wave radar evaluation board based on FMCW (frequency-modulated continuous wave) technology, configure the transmission mode as a 12T16R (12 transmitting antennas and 16 receiving antennas) mode, and specifically configure a transmission mode of "transmitting two subframes with different idle_time in a single frame period" in a single frame chirp (pulse) time, so as to meet the requirements of high resolution and improved maximum sensing speed.

[0008] Step 2, the designed signal processing system will process the signals received by the millimeter wave radar frame by frame. In each frame period, the signals of the two subframes are processed in turn, including range FFT (range dimension fast Fourier transform) radar ranging, doppler FFT (doppler dimension fast Fourier transform) radar speed measurement, CFAR (constant false alarm detection), and DOA (direction of arrival) estimation, time stamp matching optical photos, and drawing radar point cloud data graphs on the optical photos to obtain a point cloud data set generated after DOA estimation of the two subframes.

[0009] Step 3, according to the point cloud data set, the point cloud clustering data set of K0 and K1 targets in the two subframes is obtained by clustering processing.

[0010] Step 4, according to the point cloud clustering data set, the BEV (bird's eye view) information is used to match between K0 targets and K1 targets to obtain a point cloud matching data set containing matching success and matching failure.

[0011] Step 5, according to the point cloud matching data set, one of the two different error matrix schemes provided according to different application scenarios is selected to calculate and obtain the error minimum disambiguation strategy.

[0012] Step 6, according to the disambiguation strategy, the point cloud matching data set is disambiguated to obtain the disambiguated radar point cloud result, and the disambiguated radar point cloud result is drawn on the optical photo.

[0013] Compared with the prior art, the present application has the following beneficial effects: the present application solves the problem of small maximum sensing speed caused by the limitation of spatial resolution and speed resolution of the high-resolution millimeter wave radar for auxiliary intelligent driving. While ensuring the spatial and speed resolution of the millimeter wave radar, a wider sensing speed detection range is achieved at the signal processing level. Due to hardware limitations, under the premise that the sampling rate and frequency sweep speed cannot be changed, in order to increase the number of transmitting and receiving antennas while reducing the spatial resolution of the millimeter wave radar, the algorithm of the signal processing is improved and optimized to enable the millimeter wave radar to measure the speed of vehicles in a high-speed environment, thereby meeting the application requirements of the millimeter wave radar in a high-speed scenario. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1is a chirp configuration schematic (1);

[0015] Figure 2 is a chirp configuration schematic (2);

[0016] Figure 3 is a speed extension algorithm flowchart (1);

[0017] Figure 4 is a speed extension algorithm flowchart (2)

[0018] Figure 5 is a speed ambiguity phenomenon schematic diagram;

[0019] Figure 6 is a speed de-ambiguity image (1);

[0020] Figure 7 is a speed de-ambiguity image (2); DETAILED DESCRIPTION

[0021] The present application will be further described and illustrated in conjunction with the accompanying drawings and specific embodiments. The technical features of each embodiment in the present application can be combined accordingly without conflict.

[0022] The present application is based on a Texas Instruments TI MMWCAS-RF-EVM (AWR2243) and MMWCAS-DSP-EVM cascade radar kit, a high-resolution mode of 12T16R (12 transmitting antennas, 16 receiving antennas), to improve the maximum sensing speed range.

[0023] In the FMCW technology of the millimeter wave radar, specifically comprising: angle resolution In the formula: Δθ represents the angle resolution, d is the distance between the two adjacent receiving antennas, λ is the wavelength of the IF signal generated after the receiving and transmitting signals are processed by the mixer, K is the number of receiving antennas, and θ represents the angle of arrival; distance resolution In the formula: ΔR represents the distance resolution, c represents the speed of light, and B represents the scanning bandwidth of the hardware in the FMCW mode; speed resolution In the formula: N is the number of pulses transmitted by one transmitting antenna in one frame, T c is the time required for all transmitting antennas to continuously transmit a round of pulses, i.e., the frequency sweeping period, so NT c represents the frame period; maximum measurement speed

[0024] In the design of the high-resolution millimeter wave radar, in order to improve the spatial resolution of the millimeter wave radar, the MIMO method is introduced, the number of TX transmitting antennas is increased, which can equivalently and multiple times increase the number of receiving antennas N, and the frequency sweeping period T cThe maximum measurable velocity v c will also increase, i.e. a smaller spatial resolution (mainly Δθ) and a smaller velocity resolution (ΔV) can be obtained; but T max the increase of T r will result in the maximum measurable velocity v max decreasing, i.e. the maximum perception velocity of the millimeter wave radar decreases, and the perception velocity range narrows.

[0025] When the actual velocity Vr of a target is greater than the highest detection velocity V a of the radar, the velocity obtained through the doppler FFT processing will be ambiguous, and the ambiguous velocity is defined as V a The ambiguous velocity V r and the actual velocity Vr of the target satisfy the formula: V max = V a mod V r , where the ambiguous velocity V max is the remainder of the actual velocity V r modulo the highest detection velocity V a , mod represents the remainder, and V max = V 0max +k×V 1max , where k is an unknown coefficient. As shown in FIG. 1, the millimeter wave radar is set to transmit two subframes in a frame to perceive the velocity of an object, and the actual velocity of the vehicle is about 35 km / h, but since it exceeds the maximum perception range of subframe 1 (the maximum perception range is -4.950 m / s ~ 4.796 m / s) and subframe 2 (the maximum perception range is -3.322 m / s ~ 3.218 m / s), an ambiguous phenomenon appears in the dopplerMap II heat map representing the distance-velocity relationship.

[0026] A method for improving the maximum perception velocity of a high-resolution millimeter wave radar, and the specific steps are as follows:

[0027] Step 1: Configure parameters on a millimeter wave radar evaluation board (TI MMWCAS-RF-EVM (AWR2243) + MMWCAS-DSP-EVM) based on the frequency-modulated continuous wave (FMCW) technology, configure the transmission mode to be 12T16R (12 transmitting antennas and 16 receiving antennas), and configure the transmission mode to be "transmit two subframes with different idle_time in a single frame period" to meet the requirements of high resolution and improved maximum perception velocity.

[0028] As shown in FIG. 2, the single frame period of the millimeter wave radar is 100 ms, which includes two subframes, subframe 1 and subframe 2. The specific setting details are as follows: Figure 1

[0029] ​The first 15 ms subframe (subframe0) is defined as Fast-Chirp subframe. There are N=64 loops in the first 15 ms subframe. Because of the MIMO technology, 12 transmitting antennas transmit chirp in turn in each loop. Each chirp is configured with idle time idle_time0=2.00 μs. The number of sampling points of a single chirp pulse is 256. According to the formula The velocity resolution at this time is Δv0=0.1547 m / s, according to the formula The maximum perceived velocity can be calculated as v 0max =4.95 m / s.

[0030] The last 85 ms subframe (subframe1) is defined as Slow-Chirp subframe. There are N=64 loops in the last 85 ms subframe. Because of the MIMO technology, 12 transmitting antennas transmit chirp in turn in each loop. Each chirp is configured with idle time idle_time1=10.00 μs. The number of sampling points of a single chirp is 256. According to the formula The velocity resolution at this time is Δv1=0.1038 m / s, according to the formula The maximum perceived velocity can be calculated as v 1max =3.32 m / s.

[0031] The single pulse configuration parameters of the millimeter wave radar (TI MMWCAS-RF-EVM (AWR2243) + MMWCAS-DSP-EVM) are as follows Figure 2The upper left corner shows the manual configuration information, such as idel_time (idle time, which determines the sweep period, subframe 1 is 2.000000us, subframe 2 is 10.000000us), adc_start_time (ADC scan start time, subframe 1 is 2.720000us, subframe 2 is 2.720000us), numTxAnt (number of transmit antennas, subframe 1 is 12, subframe 2 is 12), num_chirps (the number of pulses transmitted by one transmit antenna in a frame, subframe 1 is 64, subframe 2 is 64), adc_samples (ADC sampling points, subframe 1 is 256, subframe 2 is 256) and other parameters based on manual setting, and the remaining parameters are automatically generated by software, such as Rmax (maximum detection distance, subframe 1 is 80.351403m, subframe 2 is 80.351403m), Rres (distance resolution, subframe 1 is 0.313873m, subframe 2 is 0.313873m), Vmax (maximum sensing speed, subframe 1 is 4.950871m / s, subframe 2 is 3.322295m / s), Vres (velocity resolution, subframe 1 is 0.154715m / s, subframe 2 is 0.103822m / s) and the like. This transmission mode is named mode3.

[0032] The calibrated, transmitted and received raw signals are a 256 (single chirp sampling points) * 64 (loop number) * 12 (transmit antenna number) * 16 (receive antenna number) four-dimensional complex signal matrix in each subframe period within each frame period. It can be seen that Figure 2 The right half of the figure shows the structure of the received raw signal of the receiving end.

[0033] Step 2, the designed signal processing system will process the signals received by the millimeter wave radar frame by frame. In each frame period, the signals of the two subframes are processed in turn, including rangeFFT radar ranging, dopplerFFT radar speed measurement, CFAR constant false alarm processing, DOA estimation, timestamp matching optical photo, and drawing radar point cloud data graph on the optical photo. The point cloud data set generated after DOA estimation of the two subframes is named angleEst0 and angleEst1 in the system.

[0034] As Figure 3 and Figure 4The method flowchart is shown, frame by frame processing, in each frame period, the original signal of two subframes: 256*64*12*16 four-dimensional complex signal matrix, dimension 1: 256 sampling points are processed by rangeFFT, then dimension 2: 64 loops (circulation) are processed by dopplerFFT radar speed, then dimension 3 and dimension 4: 12*16 receiving and transmitting antenna group data are compressed into 192 dimensions (MIMO technology), then 256*64*192 three-dimensional data are processed by CFAR (constant false alarm processing), P targets are detected, and the output is a detection_results structure, including the following information: target distance, target speed, 1*192 dimension data containing angle information corresponding to the target and target SNR (signal-to-noise ratio). Then the detection_results structure is processed by DOA estimation, and Q0 target points are estimated, and the output is an angleEst structure, that is, a point cloud data set, which contains information: target distance, target speed, target point azimuth, target point pitch angle and target SNR (signal-to-noise ratio).

[0035] The processing flow of two subframes (subframe0 and subframe1) is the same.

[0036] Step 3, according to the point cloud data sets (angleEst0 and angleEst1) generated after the DOA estimation of two subframes, the point cloud clustering data sets of two subframes containing K0 and K1 targets are obtained respectively by clustering processing.

[0037] First, the angleEst structure output by the DOA estimation of two subframes is converted to the rectangular coordinate system (xyz coordinate) according to the distance, azimuth and pitch angle of the target, and then the DBSCAN clustering is performed in the BEV graph with x coordinate, y coordinate and SNR (signal-to-noise ratio) three dimensions. The DBSCAN clustering processing includes two parameters: epsilon = 2 (distance threshold), minpts = 5 (minimum number of samples in the field). The point cloud clustering data sets of two subframes containing K0 and K1 targets are obtained respectively.

[0038] Step 4, according to the point cloud clustering data sets containing K0 and K1 targets of two subframes, the BEV information is used to match between K0 targets and K1 targets, and a point cloud matching data set containing matched and unmatched is obtained.

[0039] The point cloud clustering data sets containing K0 and K1 targets of two subframes are used to match the positions in space by using the distance and coordinates of the target.

[0040] Step 5, according to a point cloud matching data set containing matching success and unmatching success, one of two different error matrix schemes provided according to different application scenarios is selected to calculate and obtain the error minimum deblocking strategy.

[0041] The two different error matrix schemes, the first scheme is: in low speed environment, 3*3 error matrix is used, the theoretical deblocking speed interval is controlled in -9.9m / s~+9.9(m / s)(35.64km / h), which is suitable for vehicle and pedestrian detection; the second scheme is: in high speed environment, 5*5 error matrix of known orientation label (i.e. the label indicating that the object is close to or far away from the millimeter wave radar) is used, the theoretical deblocking speed interval is controlled in 0m / s~+16.61(m / s)(59.80km / h). For roadbed millimeter wave application scenario, two sets of millimeter wave radar devices can be deployed on both sides of the road for time multiplexing.

[0042] The size of the error matrix needs to be set according to the Chinese remainder theorem to calculate the theoretical maximum deblocking speed range, which in turn assists the configuration of the error matrix. The Chinese remainder theorem uses the information received twice by sampling, clusters according to the position information of each sub-frame signal, and determines whether it is the same vehicle to further restore the speed of the vehicle.

[0043] According to the Chinese remainder theorem, i.e. for equation (S) Where m1, m2,..., m n are pairwise coprime, then the value of x can be obtained through a1, a2,..., a n .

[0044] According to Figure 1 and Figure 2 , in the present application, two different T c are defined, and the V max of the two chirps corresponding to the two T r are different. The V a detected under the two chirps of the same target are also different. According to the Chinese remainder theorem (if a natural number is taken as a remainder with respect to a group of coprime factors, then under the condition of knowing the remainder and the coprime factors, the original natural number can be calculated), the V a calculated from the two chirps can be used to restore the actual speed V r of the target.

[0045] According to Figure 3 and Figure 4The algorithm flow of the same vehicle must be accumulated at least twice in a frame period (corresponding to the fast-chirp and slow-chirp configured in step 1), and after the target matching, the real speed within the allowable error range is recovered. When recovering the real speed of the target vehicle, the fuzzy speed array perceived by the two sub-frames is V1, V2.

[0046] The speed perceived by different parts of the same target will have slight differences, so the fuzzy speed exists in the form of an array. The maximum perceived speed range span corresponding to fast-chirp and slow-chirp is V scale1 = 2 x v 1max = 6.64 m / s, V scale2 = 2 x v 0max = 9.90 m / s s , where v 1max and v 0max are the maximum perceived speeds corresponding to fast-chirp and slow-chirp.

[0047] According to the two different error matrix schemes proposed at the beginning of step 5, Table 1 and Table 2 show a certain frame 3*3 error matrix for processing 20km / h single target recognition data, and a certain frame 5*5 error matrix for processing 50km / h single target recognition data with known orientation label:

[0048] Table 1: 3*3 error matrix (0.6522 is the target)

[0049] dis (error) V1-1 x V scale1 ]]> [CDATA[V1]] [CD AT] V1+1x V scale1 ]]> V2-1 x V scale2 ]]> 6.1206 15.705 108.1583 <![CDATA[V2]]> 145.5953 31.689 0.6522 V2+1 x V scale2 ]]> 469.0949 231.69 77.1711

[0050] Table 2: 5*5 error matrix with known orientation label (the vehicle is known to approach the radar center, i.e. the radial velocity is negative, so it is also a 3*3 matrix in form, and the data 1.0791 is the target)

[0051] dis (error) [CAT] V1-2 x V scale1 ]]> V1-1 x V scale1 ]]> <![CDATA[V1]]> V2 - 2 x V scale2 ]]> 73.1625 224.7138 459.1336 V2-1 x V scale2 ]]> 1.0791 29.1399 140.0694 <![CDATA[V2]]> 113.0205 17.5910 5.0300

[0052] The design of the error matrix refers to the theoretical maximum speed range of the solvable ambiguity, which is determined by the Chinese Remainder Theorem.

[0053] In the above table, dis (error) is described as: the fuzzy speed array V1, V2 perceived by the two sub-frames of fast-chirp and slow-chirp, which is transformed according to the formula (strategy) described in the first and second rows of the above matrix. The smaller the dis (error) is, the more accurate and effective the ambiguity resolution formula (strategy) is. Select the ambiguity resolution formula (strategy) with the smallest error, and apply it to the fuzzy speed array of the two groups of point clouds of fast-chirp and slow-chirp, and take the average value to obtain the real speed.

[0054] Step 6, according to the error minimum deblurring strategy, the point cloud matching dataset is deblurred to obtain the deblurred radar point cloud result, and the deblurred radar point cloud result is drawn on the optical photo.

[0055] Reference can be made to Figure 5 , Figure 6 and Figure 7 , Figure 5 is a speed ambiguity phenomenon that occurs when the maximum perception speed improvement method of the present application is not used, and the point cloud speed in the figure is the ambiguous speed. The left two figures (DopplerMap I) are the relationship between ReceivePower (return energy) and Range (distance), which is used to reflect the distance between the radar and the perception target; the middle two figures (DopplerMap II) are the heat map between Range (distance) and Velocity (speed), and the other parameter color of the heat map is Receive Power (return energy), and the white area in the figure indicates that a detection target appears in the “distance” and “speed” coordinate area; the right four figures from top to bottom represent the point cloud-photo projection graph of subframe1, the spatial point cloud distribution graph of subframe1, the point cloud-photo projection graph of subframe2, and the spatial point cloud distribution graph of subframe2 (the upper part of each figure records the time stamp of the picture or radar). It can be found from Figure 5 that when the experimental perception target is a small car traveling at a speed of 35km / h towards the radar, the millimeter wave radar has a significant speed ambiguity phenomenon in the DopplerMap. After using the maximum perception speed improvement method of the present application, the result is: Figure 6 is the deblurred point cloud-photo projection graph, and the point cloud is projected onto the optical image, Figure 7 is the deblurred spatial point cloud distribution graph, and the actual data deblurred speed is -10.23m / s, which is close to the true value.

[0056] In order to further prove the effect of the present application, a GNSS integrated inertial navigation system is installed on a single target car, and the original data obtained by GNSS is converted and extracted to latitude, longitude, north-south speed and east-west speed, which is projected onto the line connecting the car and the millimeter wave radar as the true value (Ground truth); according to the time stamp of the internal frame of the equipment, the true value (Ground truth) and the deblurred speed (Deblur) of the single target car processed by the “method for improving the maximum perception speed of high-resolution millimeter wave radar” are matched and compared, and the error (Error) is obtained, as shown in Table 3:

[0057] Table 3

[0058] true value (km / h) 4.61 13.99 17.83 25.70 35.47 45.16 method (km / h) 4.64 13.92 17.77 25.95 35.96 44.78 error (%) 0.66 0.47 0.32 1.00 1.41 0.84

[0059] The above-described embodiments are merely preferred embodiments of the present application, but are not intended to limit the present application. Those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Therefore, any technical solutions obtained by equivalent replacement or equivalent transformation shall fall within the scope of the present application.

Claims

1. A method of boosting the maximum perception speed of a high-resolution millimeter wave radar, characterized by It comprises the following steps: Step 1, configure parameters on the millimeter wave radar evaluation board based on frequency modulation continuous wave (FMCW) technology, configure the transmission mode as 12T16R mode, and configure the transmission mode of "transmitting two subframes with different idle_time in a single frame period"; The single frame period of the millimeter wave radar transmission is 100 ms, the subframe of the first 15 ms is defined as a fast pulse subframe, and the subframe of the last 85 ms is defined as a slow pulse subframe; Step 2, design a signal processing system, which will process the signals received by the millimeter wave radar frame by frame to obtain a point cloud dataset generated after DOA estimation of two subframes; Step 3, according to the point cloud dataset, respectively perform clustering processing to obtain point cloud clustering datasets containing K0 and K1 targets for two subframes; Step 4, according to the point cloud clustering dataset, use the bird's eye view (BEV) information to match between K0 targets and K1 targets, and obtain a point cloud matching dataset containing matching success and matching failure; Step 5, according to the point cloud matching dataset, select an error matrix scheme according to different application scenarios, calculate and obtain a defuzzification strategy with the smallest error; The size of the error matrix is set according to the Chinese remainder theorem to calculate the theoretical maximum defuzzification speed range, which is used to assist the configuration of the error matrix; Step 6, according to the defuzzification strategy, perform defuzzification operation on the point cloud matching dataset to obtain defuzzification radar point cloud results, and draw the defuzzification radar point cloud results on the optical photo.

2. The method of claim 1, wherein: The frame-by-frame processing of step 2 specifically includes: in each frame period, the signals of the two subframes are processed in turn by range fast Fourier transform (range FFT) radar ranging, doppler fast Fourier transform (doppler FFT) radar speed measurement, constant false alarm rate (CFAR) detection, and direction of arrival (DOA) estimation, and the optical photo is matched according to the timestamp, and the radar point cloud data graph is drawn on the optical photo.

3. The method of claim 1, wherein: The point cloud dataset of step 2 contains information such as target distance, target speed, target point azimuth, target point elevation, and target signal-to-noise ratio (SNR).

4. The method of claim 3, wherein: The clustering processing of step 3 specifically includes: Convert the target to the Cartesian coordinate system according to the distance, target point azimuth, and target point elevation, and then perform DBSCAN clustering in the bird's eye view (BEV) with x coordinate, y coordinate, and signal-to-noise ratio (SNR) as three dimensions.

5. The method of claim 1, wherein: The error matrix scheme of step 5 has two sets, which are: In a low-speed environment, a 3x3 error matrix is used to control the theoretical defuzzification speed interval to -9.9 m / s~+9.9 m / s; In a high-speed environment, a 5x5 error matrix with known orientation label is used to control the theoretical defuzzification speed interval to 0 m / s~+16.61 m / s.

Images