Point cloud correction method based on combination of point cloud association and phase offset characteristics

Abstract

The invention discloses a point cloud correction method based on combination of point cloud association and phase offset characteristics. The method comprises the following steps: step 1, sampling; 2, carrying out two-dimensional FFT (Fast Fourier Transform) on radar echoes so as to obtain distance information and speed information of a measured target; 3, carrying out incoherent accumulation: taking logarithms of two-dimensional FFT results of all antenna channels and carrying out superposition; step 4, CFAR detection is carried out; step 5, speed expansion based on Doppler phase offset compensation assumption; step 6, calculating and reporting the coordinates of the measured target; step 7, point cloud matching and correction; and step 8, performing tracking processing on the corrected target detection layer result. According to the method, speed errors and angle errors caused by a speed expansion algorithm based on Doppler phase offset compensation assumption when the consistency between the channels is poor can be recognized and corrected, and therefore the correctness of results and the correctness of follow-up processing can be guaranteed as much as possible.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing, and in particular to a point cloud correction method based on combining point cloud association and phase deviation characteristics. Background Art

[0002] Millimeter-wave radar has the advantages of high resolution, strong anti-interference ability, strong penetration, and small size, and is widely used in the fields of autonomous driving and traffic flow monitoring. Millimeter-wave radar can obtain the distance information, speed information, and angle information of the measured target by emitting and receiving electromagnetic waves.

[0003] Angular resolution, speed measurement range, and range are important criteria for measuring the performance of millimeter-wave radar. Angular resolution can be improved by increasing the number of receiving antennas. However, due to various limitations such as cost and design, the number of receiving antennas cannot be increased blindly. Time division multiplexing (TDM) multiple input multiple output (MIMO) is widely used in millimeter-wave radar because it can improve angular resolution at a low hardware cost.

[0004] However, the use of TDM MIMO has caused the maximum unambiguous speed to be reduced exponentially, making the speed measurement capability of millimeter-wave radar unable to meet the requirements of traffic neighborhoods. In addition, due to the relative motion between the target and the radar, the phase change caused by the relative motion is coupled to the receiving antenna, resulting in inaccurate angle estimation.

[0005] Therefore, when applying TDM MIMO, it is necessary to expand the speed measurement range and compensate for the phase difference caused by relative motion, so as to meet the actual speed measurement requirements and obtain correct angle estimation.

[0006] Existing technical solutions:

[0007] To solve the problem of exponentially reduced speed measurement range, the following solutions are currently used:

[0008] (1) Frequency-interference method:

[0009] The repetition frequency difference method is the most common method for resolving velocity ambiguity. It transmits multiple groups of signals, each with a different repetition frequency and mutually prime. According to the Chinese remainder theorem, the unambiguous Doppler frequency after resolution is its least common multiple, thus completing velocity resolution.

[0010] (2) Speed ​​expansion using overlapping array elements:

[0011] By adjusting the arrangement of the antennas, the virtual array elements they form overlap, and the overlapping elements do not have a phase difference caused by the spacing between the antennas. Therefore, by processing the signals of the virtual array elements, the maximum unambiguous speed is extended.

[0012] (3) Reduce the continuous wave repetition period:

[0013] The maximum unambiguous speed of millimeter waves is inversely proportional to the repetition period. By reducing the repetition period, that is, increasing the repetition frequency, the maximum unambiguous speed can be directly increased.

[0014] (4) Velocity expansion algorithm based on target tracking assumption:

[0015] This method makes use of the target's history. It first assumes that the speed obtained by the target detection layer is the fuzzy speed, then calculates the possible values ​​of the target's actual speed, and then submits these possible values ​​to the target tracking layer for processing to obtain reasonable speed values, thereby solving the problem of insufficient maximum unfuzzy speed of the target detection layer.

[0016] To compensate for the phase difference caused by relative motion, the following solutions are currently used:

[0017] (5) Compensation based on Doppler phase shift:

[0018] The phase corresponding to the speed is calculated by the speed of the measured target, and this phase is the phase difference that needs to be compensated.

[0019] (6) Compensation based on array element phase difference:

[0020] This method calculates the phase relationship between array elements, that is, calculates the average value of the phase difference between virtual channels of different receiving antennas corresponding to the same transmitting antenna and the phase difference between virtual channels of adjacent receiving antennas corresponding to different transmitting antennas, and uses these two phase differences to obtain the additional phase difference caused by the relative motion of the target, thereby completing phase compensation.

[0021] In response to the above two problems, the commonly used solutions are as follows:

[0022] (7) Based on the Doppler phase offset compensation assumption:

[0023] This solution is based on the above-mentioned "Doppler phase deviation compensation". By assuming the estimated velocity value of the measured target to be unambiguous and ambiguous, the corresponding phase is calculated according to the velocity assumption value, and the correct velocity result and angle measurement result are selected by calculating the compensation results using the corresponding phases of each assumption. This method completes velocity expansion and motion compensation at the same time.

[0024] 1. The scheme of the repetition frequency difference method has the following problems:

[0025] ① This method requires at least two sets of waveforms with different repetition frequencies to be transmitted, resulting in a low system data refresh rate, which may not meet real-time requirements.

[0026] ② The echo data corresponding to each group of waveforms need to go through the same target detection processing flow (two-dimensional FFT, CFAR), but the two-dimensional FFT and CFAR have large computational complexity, resulting in the occupation of more hardware resources.

[0027] ③ The premise of deambiguation is that the target can pass CFAR detection in both sets of data. However, in practice, for example, in a low signal-to-noise ratio environment, it is not possible to detect it every time, resulting in poor robustness.

[0028] Second, the solution of using overlapping array elements to expand speed has the following problems:

[0029] Due to the existence of overlapping array elements, the array aperture is not maximized, which leads to a decrease in angular resolution, and the speed expansion capability of using this solution alone is limited.

[0030] 3. The technical solution for reducing the continuous wave repetition period has the following problems:

[0031] This increases the difficulty of radar design and makes it extremely difficult to implement in engineering.

[0032] 4. The technical solution of the velocity expansion algorithm based on the target tracking hypothesis has the following problems:

[0033] This method makes full use of the historical information of target tracking and reduces the requirements of the target detection layer, but it increases the computational complexity of target tracking, and the response to sudden changes in target speed is affected by the tracking cycle.

[0034] 5. The technical solution based on Doppler phase shift compensation has the following problems:

[0035] ① If the speed of the measured target is ambiguous, the phase difference cannot be calculated correctly, resulting in deviation or error in the compensation result.

[0036] 6. The technical solution based on the compensation of the array element phase difference has the following problems:

[0037] ① The array is required to be arranged evenly, and speed expansion cannot be achieved at the same time.

[0038] ② If there are multiple targets at the same distance and speed, the method cannot be applied because the phase information of multiple targets is coupled.

[0039] 7. The technical solution based on the Doppler phase offset compensation assumption has the following problems:

[0040] ① Although this method can expand the speed, the speed expansion range is limited and can only be expanded to several times the current maximum unambiguous speed of the transmitting antennas. The expanded speed measurement range can generally meet most situations.

[0041] ② There are requirements for channel consistency. Poor channel consistency may lead to incorrect estimation results.

[0042] ③Since different speed assumptions need to be processed, if there are multiple targets at the same distance and the same speed, it will make it impossible to screen out the correct results. Therefore, it is suitable for scenarios with incoherent targets.

[0043] Comparing the above methods, we can see that the method "based on Doppler phase offset compensation assumption" can compensate for the phase difference caused by the relative motion of the target while completing the speed expansion, and does not need to change the radar waveform, which is very suitable for the waveform configuration of TDMMIMO. The method "speed expansion algorithm based on target tracking assumption" can further make up for the limited speed expansion range of the method "based on Doppler phase offset compensation assumption". In actual engineering, these two methods are often used in combination.

[0044] Since the method "based on the Doppler phase offset compensation assumption" has requirements for channel consistency, in practical applications it may encounter the situation where the wrong assumption results are obtained, that is, the speed and angle corresponding to the wrong assumption results are obtained, which causes deviations in the results and makes subsequent processing difficult.

[0045] In order to solve the problem of the method "based on the Doppler phase deviation compensation assumption", the present invention proposes a method for finding and correcting erroneous assumption results based on the limitation of the method being applicable to incoherent target scenarios, thereby improving the reliability of the target detection layer and enabling the subsequent target tracking layer to proceed smoothly.

[0046] In addition, time division multiplexing (TDM) multiple input multiple output (MIMO) frequency modulated continuous wave (FMCW) radar is widely used due to its advantages of low hardware cost and high angular resolution. However, this TDM MIMO method faces the following two main problems in practical applications:

[0047] ① During the antenna switching time, due to the relative motion between the target and the radar, the phase change caused by the motion is coupled to the receiving antenna, affecting the angle estimation and causing inaccurate angle estimation.

[0048] ② The use of TDM MIMO leads to a decrease in the sampling rate of the slow time dimension, which causes the radar's unambiguous speed measurement range to be reduced exponentially, making it impossible to meet speed measurement requirements in application scenarios such as traffic and vehicle-mounted.

[0049] In order to solve the above two problems, it is necessary to compensate for the phase change caused by motion and expand the maximum unambiguous velocity during signal processing. At present, the velocity expansion algorithm based on the Doppler phase offset compensation assumption is often used in TDMMIMO systems to solve the above two problems. However, in practical applications, due to channel inconsistency caused by external factors such as processing, it may lead to incorrect velocity and angle results obtained by this method, which will bring difficulties or even errors to subsequent processing. Summary of the invention

[0050] The present invention proposes a point cloud correction method based on combining point cloud association and phase deviation characteristics. The present invention provides a feasible method that can identify and correct the velocity error and angle error caused by the velocity expansion algorithm based on the Doppler phase deviation compensation assumption, thereby ensuring the correctness of the result and the correctness of subsequent processing as much as possible.

[0051] The present invention is a point cloud correction method based on combining point cloud association and phase deviation characteristics, and the method comprises the following steps:

[0052] Step 1. Sampling: Construct a multiple-input multiple-output MIMO antenna array, which contains m transmitting antennas and n receiving antennas to form m×n virtual channels; each transmitting antenna transmits a signal in turn, and the receiving antenna receives the signal at the same time, and then performs ADC sampling on the radar echo;

[0053] Step 2. Perform two-dimensional FFT on the radar echo to obtain the distance and speed information of the measured target:

[0054] Step 3. Non-coherent accumulation: Take the logarithm of the two-dimensional FFT results of each antenna channel and superimpose them;

[0055] Step 4. CFAR detection: CA-CFAR is first performed along the slow time dimension on the result after incoherent accumulation, and the amplitude of the unit under test is compared with the amplitude of the mean of the sum of the noise units on both sides to determine whether the CFAR detection condition is met. Then, SO-CFAR is performed along the fast time dimension according to the CA-CFAR detection result, and the smaller value of the sum of the amplitude of the unit under test and the noise units on both sides is compared to determine whether the CFAR detection condition is met;

[0056] Step 5. Velocity extension based on Doppler phase offset compensation assumption: The algorithm based on Doppler phase offset compensation assumption assumes that the velocity V estimated by the radar est There are two cases: unambiguous and ambiguous, and the phase deviation is calculated according to each case. Then use the phase deviation value to compensate and perform the corresponding angle dimension FFT, and select the hypothesis corresponding to the highest peak value from among the hypotheses as the result;

[0057] Step 6. Calculate and report the coordinates of the target: Based on the results obtained in step 5 and combined with the results of step 2, calculate the Cartesian coordinates of the target to obtain a point cloud. This is the end of the target detection layer, and report the information of the point cloud to the target tracking layer.

[0058] Step 7. includes ① point cloud matching and ② correction: when the target tracking layer is processed, the result of the target detection layer is corrected, which is mainly divided into two steps: point cloud matching and correction;

[0059] Step 8. Perform tracking processing on the corrected target detection layer result: After completing the correction in step 7, perform tracking processing in the target tracking layer to obtain the tracking result.

[0060] Specifically, step 2 is as follows: for the ADC sampling results of each virtual channel, firstly perform fast time dimension FFT processing, that is, perform FFT transformation on the discrete data corresponding to the echo of each frequency modulated continuous signal of each virtual channel to obtain the fast time dimension, that is, the distance dimension FFT result of the echo signal;

[0061] Then, based on the result, slow-time FFT processing is performed on each virtual channel, that is, FFT transformation is performed on the discrete data corresponding to the echo of each frequency-modulated continuous signal corresponding to each sampling point of the result after static clutter filtering of each virtual channel to obtain the new slow-time dimension of the echo, that is, the speed dimension FFT result; thereby obtaining the distance information and speed information of the measured target.

[0062] To go further, the specific steps of step 5 are as follows:

[0063] ① Get the estimated speed V of the radar est , and calculate the two cases where velocity blur occurs, as follows:

[0064] V corr =V est ±2V max

[0065] V corr Indicates the possible actual speed in the case of speed ambiguity, V max represents the maximum unambiguous speed;

[0066] ②According to V est and V corr The additional phase value caused by the target's motion is calculated in two cases as follows:

[0067]

[0068] is the phase deviation value without velocity ambiguity, is the phase deviation value in the case of velocity ambiguity;

[0069] ③ Obtain the Then, the discrete data of the virtual channels corresponding to the transmitting antennas except the first transmitting antenna are calculated as follows using the assumed To make compensation:

[0070]

[0071] Where M represents the number of transmitting antennas, and m represents the number of transmitting antennas;

[0072] ④ Perform angle-dimensional FFT on the results after compensation for each hypothesis, and calculate the amplitude of the corresponding peak value;

[0073] ⑤ Further compare the amplitudes of the peak values ​​calculated in step ④. The hypothesis corresponding to the maximum value is the correct hypothesis, and the speed and angle corresponding to the hypothesis can be obtained.

[0074] Furthermore, the point cloud allocation in step 7 is as follows:

[0075] (1) The radial distance of the point cloud is calculated as R = RangeStep × RangeIdx, where RangeStep is the distance resolution and RangeIdx is the index value of the distance unit after two-dimensional FFT. The range unit index value RangeIdx is obtained by dividing the radial distance R of the target detection layer result by the distance resolution RangeStep.

[0076] (2) The radial velocity of the point cloud is calculated as V = VStep × V ExVIdx , VStep is the speed resolution, V ExVIdx is the speed unit index value, V ExVIdx Different from RangeIdx in (1), it is a positive or negative index value and may be the index value after velocity ambiguity resolution. V is obtained by dividing the radial velocity V of the target detection result by the velocity resolution VStep. ExVIdx , and then judge |V ExVIdx | is greater than or less than half of the number of slow-time dimension FFT points to determine whether the point has completed speed expansion through step 5. If speed expansion has been performed, then the point is determined according to V ExVIdx The actual size of the target is subtracted or added to the number of slow-time FFT points to obtain the index VIdx corresponding to the speed estimated by the radar. After CFAR detection, a target usually consists of multiple points, and the RangeIdx and VIdx of these multiple points are usually continuous.

[0077] (3) Perform the first search, search the current target detection layer results in order, if |RangeIdx1-RangeIdx2| of the two points is less than or equal to 1 and VIdx1=VIdx2, where RangeIdx1 and RangeIdx2 represent the distance index values ​​of the two point clouds currently involved in the calculation, and VIdx1 and VIdx2 represent the speed index values ​​of the two point clouds currently involved in the calculation estimated by the radar, then the points that meet the above conditions are regarded as a set, and the same set points are marked with the same PointIndex to indicate the set to which the current point belongs. The purpose of the first search is to distinguish the target detection layer results only by RangeIdx;

[0078] (4) Perform a second search, searching the current target detection layer results in order. If the PointIndex of the two points is different, determine whether the two points satisfy |RangeIdx1-RangeIdx2|≤1 and |VIdx1-VIdx2|≤1 at the same time. If so, the two points are from the same target, and the PointIndex of the two points, i.e., the points in the set to which they belong, is set to the same value. After traversing all points as described above, the matching based on RangeIdx and VIdx is completed, that is, the point clouds from the same target are matched.

[0079] Furthermore, the correction in step 7 is based on the point cloud matching result, specifically: before the target tracking layer, the point cloud allocation result in step 7 and the actual measurement environment are combined to determine whether the correct hypothesis result is obtained in the target detection layer, and make corresponding corrections.

[0080] Furthermore, the correction in step 7 is specifically:

[0081] (1) searching the point clouds in each set in the point cloud allocation in step 7 in sequence, and extracting the speed information, angle information and distance information of each point cloud;

[0082] (2) Determine the speed of each point in the set and make corrections.

[0083] Furthermore, in step 7 (1), points with multiple angles at the same distance and speed need to be marked separately.

[0084] Furthermore, in step 7 (2), the following situations usually occur:

[0085] The first type: all points can obtain speed results that conform to the actual speed measurement scenario through step 5, and no correction is required;

[0086] The second type: All points cannot obtain speed results that meet the actual speed measurement scenario through step 5, such as the speed is too small, too large or opposite to the actual speed, and then the measured speed V of the wrong point is determined estFor example, if the target's true speed belongs to hypothesis 0, but the measured speed belongs to hypothesis 1 or hypothesis 2, the angle needs to be corrected accordingly according to the phase deviation relationship, and V est Subtract or add the corresponding V max multiples of;

[0087] The third type: some points can obtain speed results that conform to the actual speed measurement scenario by step 5; in this case, if the speed of the error point is similar to the second case, that is, the speed is too small or opposite to the actual speed, then it is only necessary to correct the angle of the error point to the angle of the correct point in the set according to the corresponding angle relationship between the assumptions, and then convert the corresponding V est Subtract or add the corresponding V max However, in this case, the speed at the error point and the speed at the correct point are consistent with the actual speed measurement scenario to a certain extent. For example, the speed measurement of the target is respectively close to V in hypothesis 1. max The value of V corresponding to this value in the assumption 0 est In this case, it is difficult to identify the correct speed. In this case, if the set contains three hypotheses, the angle relationship between the three hypotheses is used to determine which hypothesis the target's actual speed falls into; if the set contains only two hypotheses, and one of the two hypotheses does not appear in the actual situation, for example, hypothesis 2 appears in the point cloud result of the oncoming vehicle, then the angle relationship between the two hypotheses can be used to determine which hypothesis the target's actual speed falls into; if both hypotheses are hypotheses that may appear in the actual situation, the angle relationship cannot be used alone to determine, because the speed values ​​between the hypotheses differ by 2V. max Then, the velocities of other point clouds are combined for joint judgment to determine the correct hypothesis. After the correct hypothesis is obtained, the angle is corrected according to the angle relationship and V est Subtract or add the corresponding V max multiples of.

[0088] For the above three cases, the speed is V est The velocity threshold VelocityThre that is smaller than the minimum velocity threshold that the associated points can form a tracking unit.

[0089] The present invention has the following beneficial effects:

[0090] 1. This method is based on the usage scenario limitations of the velocity expansion algorithm based on the Doppler phase compensation assumption, and the limitations of traditional clustering methods when erroneous results caused by this algorithm appear in practical applications. A clustering method based on velocity unit and distance unit index is proposed. This method can associate point clouds from the same target even when errors occur in the point clouds.

[0091] Second, this method can identify the erroneous points and correct points caused by the velocity expansion algorithm based on the Doppler phase offset compensation assumption, and correct the velocity and angle of the erroneous points.

[0092] With a small amount of calculation, this method can still identify the point cloud from the same target when there are deviations in the angle and speed estimation of the target detection layer, and correct the erroneous point cloud, thereby optimizing the processing of the subsequent target tracking layer. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] In order to more clearly and completely describe the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings used in the technical description.

[0094] Figure 1 It is the angle dimension FFT result diagram under each hypothesis when the correct hypothesis, i.e. hypothesis 1, can be obtained;

[0095] Figure 2 It is a diagram showing the angle dimension FFT results under various assumptions when the correct assumption cannot be obtained;

[0096] Figure 3 is a flow chart of the algorithm of the present invention;

[0097] Figure 4 It is the result of method 5.1 when method 5.2 is not used for correction;

[0098] Figure 5 It is the result of method 5.1 after correction using method 5.2;

[0099] Figure 6 This is the result of the traditional point cloud association method when the correction method in 5.2 is not used;

[0100] Figure 7 This is the result of the traditional point cloud association method after correction using the method in 5.2. DETAILED DESCRIPTION

[0101] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0102] Reference Figure 3 ,The process of this method is as follows. The overall process can be divided into two levels of target ,detection layer and target tracking layer. This method is mainly applied to the target ,tracking layer.

[0103] Step 1. Sampling:

[0104] A multiple-input multiple-output MIMO antenna array is constructed, which contains m transmitting antennas and n receiving antennas to form m×n virtual channels. Each transmitting antenna transmits a signal in turn, and the receiving antenna receives the signal at the same time, and then the radar echo is sampled by ADC.

[0105] Step 2. Perform 2D FFT on the radar echo:

[0106] For the ADC sampling results of each virtual channel, fast time dimension FFT processing is first performed, that is, FFT transformation is performed on the discrete data corresponding to the echo of each frequency-modulated continuous signal of each virtual channel to obtain the fast time dimension, that is, the distance dimension FFT result of the echo signal.

[0107] Since the main observation is moving targets, the fast time dimension FFT results are subjected to static clutter filtering based on phasor mean cancellation. The purpose of this method is to eliminate stationary targets. Then, based on the results, slow time dimension FFT processing is performed on each virtual channel, that is, the discrete data corresponding to the echo of each frequency-modulated continuous signal corresponding to each sampling point of the static clutter filtering result of each virtual channel is subjected to FFT transformation to obtain the new slow time dimension, i.e., the speed dimension FFT result of the echo. Thus, the distance information and speed information of the measured target are obtained.

[0108] Step 3. Incoherent accumulation:

[0109] In order to improve the signal-to-noise ratio and make the data reasonably easy to process, the two-dimensional FFT results of each antenna channel are logarithmic and superimposed.

[0110] Step 4. CFAR detection:

[0111] The results after incoherent accumulation are first subjected to CA-CFAR along the slow time dimension. The amplitude of the unit under test is compared with the amplitude of the mean of the sum of the noise units on both sides to determine whether the CFAR detection condition is met. Then, SO-CFAR is performed along the fast time dimension based on the CA-CFAR detection result. The amplitude of the unit under test is compared with the amplitude of the sum of the noise units on both sides and the smaller value of the mean is taken to determine whether the CFAR detection condition is met.

[0112] Step 5. Velocity expansion based on Doppler phase offset compensation assumption:

[0113] Assume that the nth receiving element receives the signal S from the mth transmitting element mn (t) can be expressed as follows:

[0114]

[0115] In the above formula, 2πΔf(m-1)T C is the phase coupling term caused by the relative motion between the target and the radar when the adjacent transmitting antennas are switched, and is the phase difference caused by the position of the array element, d T is the distance between transmitting antennas, d Ris the receiving antenna spacing. The target angle is detected based on the phase of the two-dimensional FFT results corresponding to the target points on all virtual antennas. From the above formula, it can be seen that in TDM MIMO working mode, due to the Doppler effect of target motion, the virtual antennas corresponding to the subsequent transmitting antennas will be different from the virtual antenna corresponding to the first transmitting antenna except and In addition, an additional Doppler phase shift of 2πΔf(m-1)T will be generated. C , which makes the phase difference of the received signal of the virtual array element not only related to the array position, but also to the speed of the target, so the Doppler phase deviation needs to be corrected before the angle estimation to ensure the accuracy of the angle estimation. The additional phase deviation on each virtual channel can be expressed as follows:

[0116]

[0117] in It is the phase coupling term caused by the relative motion between the target and the radar during the switching time of adjacent antennas, where T C is the time of a Chirp signal, V real is the true speed of the target, C is the speed of light, f c is the frequency. So the phase deviation compensation can be performed using the following formula:

[0118]

[0119] Depend on It can be seen that only by knowing the actual speed of the target can we know the actual phase deviation. Define the target speed estimated by the radar as V est , then the phase deviation calculated using the estimated velocity is It can be expressed as:

[0120]

[0121] The target speed V estimated by the radar est and the target's true velocity V real The relationship is as follows:

[0122] V real =V est +k*V max k=…,-1,0,1,…

[0123] Where V max is the maximum unambiguous speed, Where λ is the wavelength, T c Chirp time is the duration of a continuous frequency modulation wave.

[0124] So there exists |V real|≤|V max | and |V real |>|V max |Two cases, for the former, V real =V est Available Then V est Fully compensated phase deviation For the latter, when the actual speed |V real |>|V max |, the estimated speed V of the radar est is a fuzzy value, and its relationship with the actual target speed is the same as the above formula, so its phase relationship can be expressed as follows:

[0125]

[0126] Therefore, if the speed is blurred, Compensation cannot correct phase deviation

[0127] Due to the use of TDM MIMO, the interval between Chirp signals transmitted by the same transmitting antenna is proportional to the number of transmitting antennas, that is, the interval T = n*T c , where n represents the number of transmitting antennas, V max The calculation formula shows that this value will be reduced to 1 / n under TDM MIMO. Therefore, the use of TDM MIMO not only reduces the speed measurement capability by several times, but also makes it impossible to correctly perform phase deviation compensation when the actual target speed is high due to the reduction of the maximum unambiguous speed. And used for compensation to obtain the peak value corresponding to the angle in the subsequent angle FFT processing. The algorithm based on this Doppler phase deviation compensation assumption assumes that the speed V estimated by the radar is est There are two cases: unambiguous and ambiguous, and the phase deviation is calculated according to each case. Then use the phase deviation value to compensate and perform the corresponding angle dimension FFT, and select the hypothesis corresponding to the highest peak value from among the hypotheses as the result. The specific steps are as follows:

[0128] ① Get the estimated speed V of the radar est , and calculate the two cases where velocity blur occurs:

[0129] V corr =V est ±2V max

[0130] V corr Indicates the possible actual speed in the case of speed ambiguity, V max Indicates the maximum unambiguous speed.

[0131] ②According to V est and V corr The additional phase value caused by the target's motion is calculated in two cases as follows:

[0132]

[0133] is the phase deviation value without velocity ambiguity, is the phase deviation value in the case of velocity ambiguity.

[0134] ③ Obtain the Then, the discrete data of the virtual channels corresponding to the transmitting antennas except the first transmitting antenna are calculated as follows using the assumed To make compensation:

[0135]

[0136] Where M represents the number of transmitting antennas, and m represents the number of transmitting antennas.

[0137] ④ Perform angle-dimensional FFT on the results after compensation for each hypothesis, and calculate the amplitude of the corresponding peak.

[0138] ⑤ Further compare the amplitudes of the peak values ​​calculated in step ④. The hypothesis corresponding to the maximum value is the correct hypothesis, and the speed and angle corresponding to the hypothesis can be obtained.

[0139] Step 6. Calculate and report the coordinates of the target:

[0140] According to the results obtained in step 5, combined with the results of step 2, the Cartesian coordinates of the target are calculated to obtain the point cloud. The target detection layer ends here, and the information of the point cloud is reported to the target tracking layer.

[0141] Step 7. Point cloud matching and correction:

[0142] From step 5, we can know that the additional phase deviation caused by the relative motion between the measured target and the radar can be corrected in theory. However, in actual application, due to the influence of signal-to-noise ratio, channel inconsistency, etc., the correct result may not be obtained. When the measurement is wrong, the speed may be 2 or 4 times V compared to the correct speed. max The difference will also cause the angle to deviate.

[0143] In target tracking, the four modules of association (matching the target detection layer results to the tracking unit), allocation (clustering the results of the target detection layer that cannot be associated to the tracking unit and assigning new tracking units), update (correcting the tracking according to the latest results of the target detection layer in the tracking unit), and maintenance (judging the status of the tracking unit according to different conditions) may be affected by the above deviations.

[0144] First, the angle deviation brings the following problems:

[0145] In the association module, the angle relationship between the current point and the tracking unit must first be determined. When the angle difference is greater than the set threshold, it cannot be associated with the tracking unit, which results in the point that should be associated with the tracking unit not being successfully associated due to the angle deviation. It can also be associated by raising the threshold, but this will cause the target to jitter in a large range after being processed by the update module. In the maintenance module, the failure of the point to be successfully associated with the tracking unit will also make the current tracking unit unable to meet the maintenance conditions of the current state, causing the current tracking unit to extract and end tracking. At the same time, the points that cannot be associated with the tracking unit will be clustered in the allocation module. On the one hand, due to the angle deviation, when the points that should be associated with the tracking unit but have not been successfully associated meet the conditions, a new tracking unit will be formed in the allocation module. On the other hand, due to the angle deviation, the point cloud that should be clustered cannot meet the distance difference judgment conditions and cannot be successfully clustered, which will cause confusion in the tracking results.

[0146] Secondly, the speed deviation may cause the speed of the tracking unit to be wrong, which will also lead to confusion in the tracking results.

[0147] Based on the above problems, the results of the target detection layer need to be corrected when processing the target tracking layer, including point cloud matching and correction:

[0148] Step ① Point cloud matching:

[0149] Generally, whether point clouds can be clustered is determined by judging their speed difference and distance difference. However, due to the wrong result of step 5, the simple judgment based on speed difference and distance difference cannot meet the actual situation. Considering that the method in step 5 is only applicable to incoherent scenes, and in most cases, there are very few cases where multiple targets appear in the same distance unit and speed unit, the association method proposed in this method is as follows:

[0150] (1) The radial distance of the point cloud is calculated as R = RangeStep × RangeIdx, where RangeStep is the distance resolution and RangeIdx is the index value of the distance unit after two-dimensional FFT. The range unit index value RangeIdx is obtained by dividing the radial distance R of the target detection layer result by the distance resolution RangeStep.

[0151] (2) The radial velocity of the point cloud is calculated as V = VStep × V ExVIdx , VStep is the speed resolution, V ExVIdx is the speed unit index value, V ExVIdx Different from RangeIdx in (1), it is a positive or negative index value and may be the index value after velocity ambiguity resolution. V is obtained by dividing the radial velocity V of the target detection result by the velocity resolution VStep. ExVIdx , and then judge |V ExVIdx | is greater than or less than half of the number of slow-time dimension FFT points to determine whether the point has completed speed expansion through step 5. If speed expansion has been performed, then the point is determined according to V ExVIdx The actual size of the target is subtracted or added to the number of slow-time FFT points to obtain the index VIdx corresponding to the velocity estimated by the radar. After CFAR detection, a target usually consists of multiple points, and the RangeIdx and VIdx of these multiple points are usually continuous.

[0152] (3) Perform the first search, search the current target detection layer results in order, if |RangeIdx1-RangeIdx2|≤1 and VIdx1=VIdx2, where RangeIdx1 and RangeIdx2 represent the distance index values ​​of the two point clouds currently involved in the calculation, and VIdx1 and VIdx2 represent the speed index values ​​of the two point clouds currently involved in the calculation estimated by the radar, then the points that meet the above conditions are regarded as a set, and the same set point is marked as PointIndex to indicate the set to which the current point belongs. The purpose of the first search is to distinguish the target detection layer results only by RangeIdx.

[0153] (4) Perform a second search, search the current target detection layer results in order. If the PointIndex of the two points is different, determine whether the two points satisfy |RangeIdx1-RangeIdx2|≤1 and |VIdx1-VIdx2|≤1 at the same time. If they do, the two points are from the same target, and the PointIndex of the two points, i.e., the points in the set to which they belong, is set to the same value. After traversing all points as described above, the matching based on RangeIdx and VIdx is completed, that is, the point clouds from the same target are matched.

[0154] Step ② Correction based on point cloud matching results:

[0155] From step 5, we can see that the three hypotheses They are The corresponding speeds for the three cases are V est -2V max ,V est,V est +2V max , then the phase to be compensated is the opposite of it, and the three cases are respectively assumed 0, assumed 1, and assumed 2. It can be obtained that when the correct When performing compensation calculations, the phase difference caused by the relative motion between the target and the radar can be completely compensated. When compensation is performed, there is still an additional phase difference caused by 2π. Here, the assumptions in the above content are explained by the compensation of the virtual channel corresponding to the second transmit antenna of TDMMIMO with three transmit antennas and four receive antennas:

[0156] For the case of three transmit and four receive, the value to be compensated for the second transmit antenna is Assume that the actual speed meets the assumption 0(V real =V est -2V max )

[0157] Assume that the additional phase deviation of the virtual channel corresponding to the second transmitting antenna of 0 is Assume that 0 corresponds to After compensation, the additional phase deviation is corrected; using the corresponding assumption 1 Compensation At this time there are still The additional phase deviation is not compensated; the corresponding Compensation At this time there are still The additional phase deviation is not compensated. Similarly, when the actual speed of the measured target is other assumptions and the virtual channel compensation corresponding to the third transmitting antenna also conforms to the above rules. The above additional phase deviation is manifested as a left or right frequency shift of the wrong assumption peak compared to the correct assumption peak during FFT processing.

[0158] As shown in the attached figure Figure 1 and Figure 2 As shown, it represents the angle dimension FFT result diagram under different assumptions when measuring a moving-away target. Figure 1 It indicates that the angle dimension FFT result diagram under each hypothesis when the correct hypothesis, i.e. hypothesis 1, can be obtained. Figure 2 It represents the angle-dimensional FFT result diagram under various assumptions when the correct assumption cannot be obtained, where the horizontal axis is the number of FFT points and the vertical axis is the amplitude. Figure 1 In the case of the correct hypothesis, its peak has a higher amplitude, and the frequency points corresponding to the other two incorrect hypotheses are located on both sides of the frequency point corresponding to the peak of the correct hypothesis; Figure 2The amplitude of the correct assumption is lower than that of the wrong assumption 2 due to the inconsistency of amplitude and phase between channels and the signal-to-noise ratio. Although assumption 0 produces distortion, the frequency points corresponding to the peak values ​​remain on both sides of the correct assumption. At the same time, the difference between the frequency point of the peak value of the wrong assumption and the frequency point of the peak value of the correct assumption is approximately a fixed value based on the number of points of the angle-dimensional FFT. This value can be converted to obtain the angle difference under each actual assumption.

[0159] Based on the above, the result of step 7① and the actual measurement environment can be combined before the target tracking layer to determine whether the correct hypothesis result is obtained in the target detection layer, and make corresponding corrections.

[0160] (1) Search the point clouds in each set of step ① in step 7 in turn, and extract the speed information, angle information, and distance information of each point cloud. There may be multiple angles at the same distance and speed. This may be due to the presence of multiple targets in the same distance unit and the same speed unit or the failure to correct the additional phase deviation, resulting in the second peak of the angle FFT result meeting specific conditions and being regarded as the target output. Therefore, the results of multiple points at the same distance and the same speed cannot be used to correct other points. Such points need to be marked separately.

[0161] (2) Determine the speed of each point in the set and make corrections. Usually there are the following situations:

[0162] The first type: All points can obtain speed results that conform to the actual speed measurement scenario through step 5, and no correction is required.

[0163] The second type: All points cannot obtain speed results that meet the actual speed measurement scenario through step 5, for example, the speed is too small, too large, or opposite to the actual speed. Then the measured speed V of the wrong point is determined. est For example, if the target's true speed belongs to hypothesis 0, but the measured speed belongs to hypothesis 1 or hypothesis 2, the angle needs to be corrected accordingly according to the phase deviation relationship, and V est Subtract or add the corresponding V max multiples of.

[0164] The third type: some points can obtain speed results that conform to the actual speed measurement scenario by step 5. In this case, if the speed of the error point is similar to the second case, that is, the speed is too small or opposite to the actual speed, then it is only necessary to correct the angle of the error point to the angle of the correct point in the set according to the corresponding angle relationship between the assumptions, and then convert the corresponding V est Subtract or add the corresponding V max However, in this case, the speed at the error point and the speed at the correct point are consistent with the actual speed measurement scenario to a certain extent. For example, the speed measurement of the target is respectively close to V in hypothesis 1. max The value of V corresponding to this value in the assumption 0 estIn this case, it is difficult to identify the correct speed. For this case, if the set contains three hypothetical results, the angle relationship between the three hypotheses is used to determine which hypothesis the target's actual speed falls into; if the set contains only two hypotheses, and there are hypotheses in the two hypotheses that will not appear in actual situations, for example, hypothesis 2 appears in the point cloud result of the oncoming vehicle, then the angle relationship between the two hypotheses can be used to determine which hypothesis the target's actual speed falls into; if both hypotheses are hypotheses that may appear in actual situations, it cannot be determined only by the angle relationship, because the speed values ​​between the hypotheses differ by 2V max Then, the velocities of other point clouds are combined for joint judgment to determine the correct hypothesis. After the correct hypothesis is obtained, the angle is corrected according to the angle relationship and V est Subtract or add the corresponding V max multiples of.

[0165] For the above three cases, the speed is V est The velocity threshold VelocityThre that is smaller than the minimum velocity threshold that the associated points can form a tracking unit.

[0166] Step 8. Perform tracking processing on the corrected target detection layer results.

[0167] After completing the correction in step 7, tracking processing is performed in the target tracking layer to obtain a tracking result.

[0168] 1. This method is based on the usage scenario limitations of the velocity expansion algorithm based on the Doppler phase compensation assumption, and the limitations of traditional clustering methods when erroneous results caused by this algorithm appear in practical applications. A clustering method based on velocity unit and distance unit index is proposed. This method can associate point clouds from the same target even when errors occur in the point clouds.

[0169] Second, this method can identify the erroneous points and correct points caused by the velocity expansion algorithm based on the Doppler phase offset compensation assumption, and correct the velocity and angle of the erroneous points.

[0170] With a small amount of calculation, this method can still identify the point cloud from the same target when there are deviations in the angle and speed estimation of the target detection layer, and correct the erroneous point cloud, thereby optimizing the processing of the subsequent target tracking layer.

[0171] The following is a comparison of the clustering method of this method before and after point cloud correction compared with the traditional clustering method when the target detection layer measurement results show erroneous measurements caused by step 5, and the correction effect of this method on the point cloud:

[0172] The above figure shows the point cloud result of an approaching vehicle, where the horizontal axis is the horizontal distance of the point cloud relative to the radar normal, and the vertical axis is the vertical distance of the point cloud from the radar transmitting surface.

[0173] Figure 4 and Figure 6 The midpoint cloud is the result without correction by method 5.2, i.e. step ② in step 7. It can be seen that the maximum difference in lateral distance can exceed 10m, which is unacceptable in practical applications. Figure 4 It shows that the correlation method in step ① of method 5.1, i.e. step 7, is better than Figure 6 The effect of the traditional method is shown in Figure 5.1. The association method in step 7 can still identify and associate point clouds from the same target even if errors occur in the point cloud. Figure 4 The four points in the middle are all of the same color and number 2; Figure 6 It shows the result of the traditional method. When there is an error in the point cloud, the points from the same target cannot be associated. It can be seen that the four points are divided into three colors, namely 1, 2, and 3.

[0174] Figure 5 and Figure 7 The point cloud distribution of is the result after correction using method 5.2, i.e. step ② in step 7. Figure 4 and Figure 6 You can know in Figure 4 and Figure 6 The two erroneous points located at about 5m respectively in the image are corrected, so that the lateral distance no longer has an error that is unacceptable in practice, that is, the point cloud from the same target becomes concentrated. Figure 5 and Figure 7 The results are as follows: after correction by using method 5.2 (step ② in step 7), the point cloud is associated with method 5.1 and the traditional method. It can be seen that after correction, the point cloud is Figure 7 The four points in are all judged to be from the same target, that is, represented by the same color and the same number 1. As can be seen from the above four figures, this method has a good improvement effect.

[0175] Millimeter-wave radar has the advantages of high resolution, strong anti-interference ability, strong penetration, and small size, and is widely used in the fields of automatic assumption and traffic flow monitoring. Millimeter-wave radar can obtain the distance information, speed information, and angle information of the measured target by emitting and receiving electromagnetic waves.

[0176] Angular resolution, speed measurement range, and effective distance are important criteria for measuring millimeter wave radars. Angular resolution can be improved by increasing the number of receiving antennas. However, due to various limitations such as cost and design, the number of receiving antennas cannot be increased blindly. Time division multiplexing (TDM) multiple input multiple output (MIMO) is widely used in millimeter wave radars because it can improve angular resolution at a low hardware cost. However, the use of TDM MIMO has caused the maximum unambiguous speed to be reduced exponentially, and the phase change caused by the relative motion between the measured target and the radar is coupled to the receiving antenna. The above problems can be solved by a speed expansion algorithm based on the Doppler phase deviation compensation assumption, but this method is affected by the consistency between channels, signal-to-noise ratio, etc., and may cause erroneous results, thereby causing difficulties in target tracking layer processing. The method provided by the present invention makes the results of the speed expansion algorithm based on the Doppler phase deviation compensation assumption more reliable, which is conducive to subsequent processing.

Claims

1. A point cloud correction method based on combining point cloud association and phase deviation characteristics, characterized in that: The method comprises the following steps: Step 1. Sampling: construct a multiple-input multiple-output MIMO antenna array, which contains m transmitting antennas and n receiving antennas to form m×n virtual channels; Each transmitting antenna transmits the signal in turn, and the receiving antenna receives the signal at the same time, and then performs ADC sampling on the radar echo; Step 2. Perform two-dimensional FFT on the radar echo to obtain the distance and speed information of the measured target: Step 3. Non-coherent accumulation: Take the logarithm of the two-dimensional FFT results of each antenna channel and superimpose them; Step 4. CFAR detection: CA-CFAR is first performed along the slow time dimension on the result after incoherent accumulation, and the amplitude of the unit under test is compared with the amplitude of the mean of the sum of the noise units on both sides to determine whether the CFAR detection condition is met. Then, SO-CFAR is performed along the fast time dimension according to the CA-CFAR detection result, and the smaller value of the sum of the amplitude of the unit under test and the noise units on both sides is compared to determine whether the CFAR detection condition is met; Step 5. Velocity extension based on Doppler phase offset compensation assumption: The algorithm based on Doppler phase offset compensation assumption assumes that the velocity V estimated by the radar est There are two cases: unambiguous and ambiguous, and the phase deviation is calculated according to each case. Then use the phase deviation value to compensate and perform the corresponding angle dimension FFT, and select the hypothesis corresponding to the highest peak value from among the hypotheses as the result; Step 6. Calculate and report the coordinates of the target: Based on the results obtained in step 5 and combined with the results of step 2, calculate the Cartesian coordinates of the target to obtain a point cloud. This is the end of the target detection layer, and report the information of the point cloud to the target tracking layer. Step 7. includes ① point cloud matching and ② correction: when the target tracking layer is processed, the result of the target detection layer is corrected, which is mainly divided into two steps: point cloud matching and correction; Step 8. Perform tracking processing on the corrected target detection layer result: After completing the correction in step 7, perform tracking processing in the target tracking layer to obtain the tracking result.

2. The point cloud correction method based on combining point cloud association and phase deviation characteristics according to claim 1, characterized in that: Step 2 is specifically as follows: for the ADC sampling results of each virtual channel, first perform fast time dimension FFT processing, that is, perform FFT transformation on the discrete data corresponding to the echo of each frequency modulated continuous signal of each virtual channel to obtain the fast time dimension, that is, the distance dimension FFT result of the echo signal; Then, based on the result, slow-time FFT processing is performed on each virtual channel, that is, FFT transformation is performed on the discrete data corresponding to the echo of each frequency-modulated continuous signal corresponding to each sampling point of the result after static clutter filtering of each virtual channel to obtain the new slow-time dimension of the echo, that is, the speed dimension FFT result; thereby obtaining the distance information and speed information of the measured target.

3. The point cloud correction method based on combining point cloud association and phase deviation characteristics according to claim 1, characterized in that: The specific steps of step 5 are as follows: ① Get the estimated speed V of the radar est , and calculate the two cases where velocity blur occurs, as follows: V corr =V est ±2V max V corr Indicates the possible actual speed when speed ambiguity occurs, V max represents the maximum unambiguous speed; ②According to V est and V corr The additional phase value caused by the target's motion is calculated in two cases as follows: is the phase deviation value without velocity ambiguity, is the phase deviation value in the case of velocity ambiguity; ③ Obtain the Then, the discrete data of the virtual channels corresponding to the transmitting antennas except the first transmitting antenna are calculated as follows using the assumed To make compensation: Where M represents the number of transmitting antennas, and m represents the number of transmitting antennas; ④ Perform angle-dimensional FFT on the results after compensation for each hypothesis, and calculate the amplitude of the corresponding peak value; ⑤ Further compare the amplitudes of the peak values ​​calculated in step ④. The hypothesis corresponding to the maximum value is the correct hypothesis, and the speed and angle corresponding to the hypothesis can be obtained.

4. The group tracking algorithm based on the integration of base clustering and tracking functions according to claim 1 is characterized in that: In step 7, ① point cloud allocation is specifically as follows: (1) The radial distance of the point cloud is calculated as R = RangeStep × RangeIdx, where RangeStep is the distance resolution and RangeIdx is the index value of the distance unit after two-dimensional FFT. The range unit index value RangeIdx is obtained by dividing the radial distance R of the target detection layer result by the distance resolution RangeStep. (2) The radial velocity of the point cloud is calculated as V = VStep × V ExVIdx , VStep is the speed resolution, V ExVIdx is the speed unit index value, V ExVIdx Different from RangeIdx in (1), it is a positive or negative index value and may be the index value after velocity ambiguity resolution. V is obtained by dividing the radial velocity V of the target detection result by the velocity resolution VStep. ExVIdx , and then judge |V ExVIdx | is greater than or less than half of the number of slow-time dimension FFT points to determine whether the point has completed speed expansion through step 5. If speed expansion has been performed, then the point is determined according to V ExVIdx The actual size of the point cloud is subtracted or added to the number of slow-time dimension FFT points to obtain the index VIdx corresponding to the velocity estimated by the radar for the point cloud; After CFAR detection, a target usually consists of multiple points, and the RangeIdx and VIdx of these multiple points are usually continuous; (3) Perform the first search, search the current target detection layer results in order, if |RangeIdx1-RangeIdx2| of the two points is less than or equal to 1 and VIdx1=VIdx2, where RangeIdx1 and RangeIdx2 represent the distance index values ​​of the two point clouds currently involved in the calculation, and VIdx1 and VIdx2 represent the speed index values ​​of the two point clouds currently involved in the calculation estimated by the radar, then the points that meet the above conditions are regarded as a set, and the same set points are marked with the same PointIndex to indicate the set to which the current point belongs. The purpose of the first search is to distinguish the target detection layer results only by RangeIdx; (4) Perform a second search, searching the current target detection layer results in order. If the PointIndex of the two points is different, determine whether the two points satisfy |RangeIdx1-RangeIdx2|≤1 and |VIdx1-VIdx2|≤1 at the same time. If so, the two points are from the same target, and the PointIndex of the two points, i.e., the points in the set to which they belong, is set to the same value. After traversing all points as described above, the matching based on RangeIdx and VIdx is completed, that is, the point clouds from the same target are matched.

5. The point cloud correction method based on combining point cloud association and phase deviation characteristics according to claim 4, characterized in that: Correction ② in step 7 is based on the point cloud matching result, specifically: before the target tracking layer, the point cloud allocation result described in step 7 and the actual measurement environment are combined to determine whether the correct hypothesis result is obtained in the target detection layer, and make corresponding corrections.

6. The point cloud correction method based on combining point cloud association and phase deviation characteristics according to claim 5, characterized in that: The correction in step 7 is as follows: (1) searching the point clouds in each set in the point cloud allocation in step 7 in sequence, and extracting the speed information, angle information and distance information of each point cloud; (2) Determine the speed of each point in the set and make corrections.

7. The point cloud correction method based on combining point cloud association and phase deviation characteristics according to claim 6, characterized in that: In step 7 (1), points with multiple angles at the same distance and speed need to be marked separately.

8. The point cloud correction method based on combining point cloud association and phase deviation characteristics according to claim 1, characterized in that: In step 7 (2), the following situations usually occur: The first type: all points can obtain speed results that conform to the actual speed measurement scenario through step 5, and no correction is required; The second type: All points cannot obtain speed results that meet the actual speed measurement scenario through step 5, such as the speed is too small, too large or opposite to the actual speed, and then the measured speed V of the wrong point is determined est For example, if the target's true speed belongs to hypothesis 0, but the measured speed belongs to hypothesis 1 or hypothesis 2, the angle needs to be corrected accordingly according to the phase deviation relationship, and V est Subtract or add the corresponding V max multiples of; The third type: some points can obtain speed results that conform to the actual speed measurement scenario by step 5; in this case, if the speed of the error point is similar to the second case, that is, the speed is too small or opposite to the actual speed, then it is only necessary to correct the angle of the error point to the angle of the correct point in the set according to the corresponding angle relationship between the assumptions, and then convert the corresponding V est Subtract or add the corresponding V max However, in this case, the speed at the error point and the speed at the correct point are consistent with the actual speed measurement scenario to a certain extent. For example, the speed measurement close to the target is respectively close to V in hypothesis 1. max The value of V corresponding to this value in the assumption 0 est In this case, it is difficult to identify the correct speed. In this case, if the set contains three hypotheses, the angle relationship between the three hypotheses is used to determine which hypothesis the target's actual speed falls into. If the set contains only two hypotheses, and one of the two hypotheses does not appear in the actual situation, such as hypothesis 2 appears in the point cloud result of the oncoming vehicle, the angle relationship between the two hypotheses can be used to determine which hypothesis the target's actual speed falls into. If both hypotheses are hypotheses that may appear in the actual situation, the angle relationship cannot be used alone to determine the speed, because the speed values ​​between the hypotheses differ by 2V. max Then a joint judgment is made in combination with the velocities of other point clouds to determine the correct hypothesis; After getting the correct assumption, correct the angle according to the angle relationship and set V est Subtract or add the corresponding V max multiples of; For the above three cases, the speed is V est The velocity threshold VelocityThre that is smaller than the minimum velocity threshold that the associated points can form a tracking unit.

Images