Millimeter wave radar point cloud speed decoupling method suitable for steering scene

Abstract

The invention relates to the technical field of vehicle-mounted radar signal processing, in particular to a millimeter-wave radar point cloud speed decoupling method suitable for a steering scene, and the method comprises the steps: firstly obtaining point cloud data with a radial distance, a radial relative speed and a horizontal / pitching angle in a vehicle steering scene, estimating the straight speed of a vehicle through employing a formula, estimating the vcar, and carrying out the calculation of the vcar after the vcar is estimated, the influence of the point cloud on the vehicle speed is filtered out, the abnormal speed at the position of the large horizontal observation angle caused by the filtered straight speed and residual steering speed of the vehicle is obtained, at the moment, the point cloud shows a linear relation on a sine value-relative speed relation graph, the linear speed value caused by steering of the vehicle is filtered out through a linear regression or RANSAC method, and the obtained slope k contains the angular speed; finally, the obtained road edge is a static point of which the speed is 0, the vehicle has the speed, and the point cloud becomes input data of a subsequent automatic driving perception algorithm module. Therefore, the problem of inaccurate point cloud absolute speed estimation during steering is solved.

Description

Technical Field

[0001] This invention relates to the field of vehicle-mounted radar signal processing technology, and in particular to a millimeter-wave radar point cloud velocity decoupling method suitable for steering scenarios. Background Technology

[0002] Currently, automotive 4D millimeter-wave radar is typically placed in the center of the vehicle bumper. It senses the external environment by emitting and receiving electromagnetic waves. External objects act as radar reflection points, displayed as point clouds on the radar device. Due to the multi-chirp waveform configuration of millimeter-wave radar, it has Doppler velocity measurement capabilities, enabling the measurement of the relative radial velocity of the point cloud. When the vehicle is traveling in a straight line at a speed of V_car, the measured velocity of the stationary point cloud is the negative of the vehicle's speed, i.e., -V_car. In this case, only the vehicle's speed is needed to calculate the true absolute velocity of the point cloud. However, when the vehicle turns, the point cloud contains not only the vehicle's speed but also the angular velocity from the turn. Traditional methods that only filter out the vehicle speed cannot solve this problem, resulting in residual angular velocities.

[0003] Prior art related to this invention: The invention is titled "A Vehicle Speed ​​Estimation Method Based on Vehicle-Mounted Millimeter-Wave Radar," application publication number CN111308458, patent holder Beijing Ligong Ruixing Electronic Technology Co., Ltd. This technical solution is as follows: The radar received signal undergoes signal processing to obtain a range-Doppler spectrum (RD spectrum); based on the Doppler effect, the velocity value of each point cloud is directly read from the RD spectrum; based on the observation angle of the point cloud, its relative velocity along the normal direction is calculated using the formula: V_ego=V_r / cos(β+θ); using V_ego as a feature, all point clouds are clustered, and the top three clusters are selected; if three are not selected, all are selected; a quantity threshold ε is set, and the average value of clusters containing more than ε point clouds is calculated, and the negative sign is used to obtain the vehicle speed. The disadvantages of this existing technology are: First, it only considers the estimation of the vehicle's speed and does not estimate the steering angular momentum, so it cannot solve the problem of accurately estimating the absolute speed of the point cloud during steering; Second, it does not refer to the previous frame for correction, which will lead to incorrect estimation of the vehicle's speed when the point cloud is sparse, resulting in a large area of ​​false targets. Summary of the Invention

[0004] The purpose of this invention is to provide a millimeter-wave radar point cloud velocity decoupling method suitable for turning scenarios, aiming to solve the problem of inaccurate point cloud absolute velocity estimation during turning.

[0005] To achieve the above objectives, this invention provides a millimeter-wave radar point cloud velocity decoupling method suitable for turning scenarios, comprising the following steps:

[0006] The radar echoes are processed to obtain point clouds with velocity, range, and angle information;

[0007] According to the formula, the radial relative velocity of the point cloud is calculated based on the pitch angle and the horizontal angle;

[0008] The vehicle speed is obtained by taking the average value of the points near the 0° angle.

[0009] The obtained vehicle speed is compared with the previous two frames. If the error is greater than ε, the estimated value of the current frame is discarded and the result of the previous frame is used. If there are no previous two frames, the vehicle speed of 10 frames is calculated first, outliers are removed and then the average value is taken as the initial vehicle speed of the previous two frames.

[0010] By filtering out vehicle speed using a formula, only the velocity component caused by angular velocity remains. Its value is linearly related to the sine of the horizontal observation angle. The slope and slant distance of the straight line are obtained by using linear regression or RANSAC method. The slope value divided by 4 times the vehicle length is the yaw rate.

[0011] The true absolute velocity of the point cloud is obtained by filtering out the vehicle's angular velocity from the acquired slope and slant distance.

[0012] The step of "processing radar echoes to obtain point clouds with velocity, range, and angle information" includes the following steps:

[0013] The 4D millimeter-wave radar transmits a linear frequency modulated (FMCW) signal. The electromagnetic wave is reflected back to the radar receiver after being reflected by the actual object. After mixing, an intermediate frequency (IF) signal is obtained in the fast time dimension. The radial distance of the target is obtained by performing FFT processing on the mixed signal along the range direction.

[0014] The Doppler frequency is obtained by performing an FFT in the slow time dimension, and then the radial relative velocity of the target is obtained. The angle is calculated by using the digital beamforming (DBF) method to obtain the distance, velocity and angle information of the point cloud.

[0015] In the section "Calculate the radial relative velocity of the point cloud based on the pitch angle and horizontal angle according to the formula", the formula is:

[0016]

[0017] Where f d f is the Doppler frequency; c is the speed of light; f c This is the carrier frequency for the transmitted signal.

[0018] In the section "taking the average of point cloud data near 0° angle to obtain the vehicle speed", the radial relative velocity of the point cloud acquired by the millimeter-wave radar placed at the front of the vehicle can be expressed as follows:

[0019]

[0020] Where vcar is the vehicle's straight-line velocity along the 0° normal, vm is the velocity component of the moving target along the 0° normal, ω is the vehicle's steering angular velocity, d is the length from the rear axle center of the vehicle to the radar, θ is the horizontal observation angle, and φ is the pitch observation angle; the Doppler velocity measurement result of the stationary point cloud at the 0° normal position is the vehicle's speed. Therefore, it can be obtained by taking the point cloud in the range of [-5°, 5°], removing the maximum and minimum outliers, and then taking the average, i.e.:

[0021]

[0022] Where vk is the radial relative velocity of the point cloud within the interval, and n is the number of point clouds within the interval, the vehicle speed can be obtained through the above formula.

[0023] In the section "After filtering out vehicle speed using a formula, only the velocity component caused by angular velocity remains. Its value is linearly related to the sine of the horizontal observation angle. The slope and slant distance of the straight line are obtained using linear regression or RANSAC methods. The slope value divided by 4 times the vehicle length is the yaw rate," the radial relative velocity of the point cloud after filtering out the vehicle speed in this frame is:

[0024] .

[0025] This invention discloses a millimeter-wave radar point cloud velocity decoupling method applicable to steering scenarios. In a vehicle steering scenario, point cloud data with radial distance, radial relative velocity, and horizontal / pitch angles is first acquired. The vehicle's straight-line velocity (v_car) is estimated using a formula. After estimating v_car, the influence of vehicle speed in the point cloud is filtered out, resulting in the filtered vehicle straight-line velocity. Residual steering velocity causes velocity anomalies at large horizontal observation angles. At this point, the point cloud exhibits a linear relationship on the sine-relative velocity graph. Linear regression or RANSAC methods are used to filter out the linear velocity values ​​caused by vehicle steering. The calculated slope k includes angular velocity, ultimately yielding static points along the roadside with zero velocity, while the vehicle retains its velocity. This type of point cloud becomes the input data for subsequent autonomous driving perception algorithm modules. This solves the problem of inaccurate absolute velocity estimation of point clouds during steering. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the phase difference of the array's angle of arrival.

[0028] Figure 2 This is a schematic diagram of the process for filtering out the straight-line speed and steering angular velocity of a vehicle.

[0029] Figure 3 This is a schematic diagram of the original point cloud velocity; where a represents the two-dimensional point cloud distribution when viewed from above; and b represents the relationship between point cloud velocity and the sine of the horizontal angle.

[0030] Figure 4 This is a schematic diagram of the point cloud after filtering out the speed of straight-moving vehicles; where a is the two-dimensional point cloud distribution when viewed from above; and b is a graph showing the relationship between point cloud speed and the sine of the horizontal angle.

[0031] Figure 5 This is a schematic diagram of the absolute velocity of the point cloud; where a is the two-dimensional point cloud distribution when viewed from above; and b is a graph showing the relationship between the point cloud velocity and the sine of the horizontal angle.

[0032] Figure 6 This is a flowchart of a millimeter-wave radar point cloud velocity decoupling method suitable for turning scenarios provided by the present invention. Detailed Implementation

[0033] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0034] Please see Figures 1 to 6 This invention provides a millimeter-wave radar point cloud velocity decoupling method suitable for turning scenarios, comprising the following steps:

[0035] S1 processes the radar echo to obtain a point cloud with velocity, range, and angle information;

[0036] The 4D millimeter-wave radar transmits a linear frequency modulated (FMCW) signal. The electromagnetic wave is reflected back to the radar receiver after being reflected by the actual object. After mixing, an intermediate frequency (IF) signal is obtained in the fast time dimension. The radial distance of the target is obtained by performing FFT processing on the mixed signal along the range direction.

[0037] The Doppler frequency is obtained by performing an FFT in the slow time dimension, and then the radial relative velocity of the target is obtained. The angle is calculated by using the digital beamforming (DBF) method to obtain the distance, velocity and angle information of the point cloud.

[0038] Specifically, 4D millimeter-wave radar typically transmits a linear frequency modulated continuous wave (FMCW) signal. Because the frequency of the FMCW signal increases over time, similar to birdsong, it is also called chirp.

[0039] A vehicle-mounted millimeter-wave radar transmits 256 or 512 chirs at a time. The time variation within a single chirp is called "fast time," and the time variation between multiple chirs is called "slow time." After the electromagnetic waves are reflected by the actual object, they return to the radar receiver, where they are mixed to obtain the intermediate frequency signal in the fast time dimension.

[0040]

[0041] Where σ represents the transmit / receive gain, transmit power, and coupling amount between the reflector's RCS (radar scattering interface) and propagation attenuation; j is the imaginary unit; fc is the transmit signal carrier frequency, typically 77GHz to 79GHz; fk is the frequency modulation slope; τ is the transmit / receive delay; t is time; and fkτ²² is the residual video phase, whose secondary perturbation is generally small and negligible. By performing FFT processing on the mixed signal along the range direction, the radial range of the target can be obtained as follows:

[0042]

[0043] S2 calculates the radial relative velocity of the point cloud based on the pitch angle and horizontal angle according to the formula;

[0044] The formula is:

[0045]

[0046] Where f d f is the Doppler frequency; c is the speed of light; f c This is the carrier frequency for the transmitted signal.

[0047] Specifically, after range compression following mixing, a single chirp signal transforms into multiple aliased monotone signals in the fast time dimension, with each monotone frequency representing range information. The target's velocity information is reflected in the phase change in the slow time dimension. The target's radial relative velocity is:

[0048]

[0049] Where fd is the Doppler frequency, which can be obtained by performing an FFT along the slow time dimension after completing the Fast Fourier Transform (FFT).

[0050] After estimating the target range and velocity, the peak point where the target exists can be selected on the RD spectrum by using the peak search method for angle calculation. The two-dimensional angle measurement of millimeter-wave radar generally adopts the digital beamforming (DBF) method. Taking a one-dimensional linear array as an example, as shown in Figure 1.

[0051] Taking array element 1 as a reference, the arrival time of the incoming wave signal at array element 2 is ahead of the arrival time at array element 1 by τ = dsinθc, and so on, the arrival time of the incoming wave at array element N is ahead of the arrival time at array element 1 by (N−1)τ. At this time, the array echo signal can be expressed as:

[0052]

[0053] Therefore, a steering matrix can be designed to traverse all observation angles, and multiplied by the conjugate of the echo signal to obtain the spatial spectrum:

[0054]

[0055] The spatial spectrum vector Y(θ) is obtained from the above formula. At this time, there is a peak in the angle of the target's direction of arrival due to the coherent accumulation effect. The angle value can be obtained by searching for the peak. The horizontal and pitch angles are obtained in the same way. Thus, the distance, velocity and angle information of the point cloud are obtained.

[0056] S3 takes the average value of the points near the 0° angle and calculates the speed of the vehicle.

[0057] The radial relative velocity of the point cloud acquired by a millimeter-wave radar placed at the front of the vehicle can be expressed as follows:

[0058]

[0059] Where vcar is the vehicle's straight-line velocity along the 0° normal, vm is the velocity component of the moving target along the 0° normal, ω is the vehicle's steering angular velocity, d is the length from the rear axle center of the vehicle to the radar, θ is the horizontal observation angle, and φ is the pitch observation angle; the Doppler velocity measurement result of the stationary point cloud at the 0° normal position is the vehicle's speed. Therefore, it can be obtained by taking the point cloud in the range of [-5°, 5°], removing the maximum and minimum outliers, and then taking the average, i.e.:

[0060]

[0061] Where vk is the radial relative velocity of the point cloud within the interval, and n is the number of point clouds within the interval, the vehicle speed can be obtained through the above formula.

[0062] Specifically, the radial relative velocity of the point cloud acquired by the millimeter-wave radar placed at the front of the vehicle can be expressed as follows:

[0063]

[0064] Where vcar is the vehicle's straight-line velocity along the 0° normal, vm is the velocity component of the moving target along the 0° normal, ω is the vehicle's steering angular velocity, d is the length from the rear axle center of the vehicle to the radar, θ is the horizontal observation angle, and φ is the pitch observation angle. From the above formula, it can be seen that the Doppler velocity measurement result of the stationary point cloud at the 0° normal position is the vehicle's speed. Therefore, it can be obtained by taking the point cloud within the range of [-5°, 5°], removing the maximum and minimum outliers, and then averaging the results.

[0065]

[0066] Where vk is the radial relative velocity of the point cloud within the interval, and n is the number of point clouds within the interval, the vehicle speed can be obtained through the above formula.

[0067] S4 compares the obtained vehicle speed with the previous two frames. If the error is greater than ε, the estimated value of the current frame is discarded and the result of the previous frame is used. If there are no previous two frames, the vehicle speed of 10 frames is calculated first, outliers are removed and then the average value is taken as the initial vehicle speed of the previous two frames.

[0068] Specifically, after obtaining the vehicle speed vcar, it is compared with the previous two frames. If the error is greater than ε, the estimated value of the current frame is discarded and the result of the previous frame is used. If there are no previous two frames, the vehicle speed of 10 frames is calculated first according to steps S1 to S3, outliers are removed and then the average value is taken as the initial vehicle speed of the previous two frames. Generally, millimeter-wave radar will perform detection with a period of 50ms after it is turned on, so the time occupied by 10 frames is about 0.5s, which is within a reasonable range.

[0069] S5 filters out vehicle speed using a formula, leaving only the velocity component caused by angular velocity. Its value is linearly related to the sine of the horizontal observation angle. The slope and slant distance of the straight line are obtained by using linear regression or RANSAC method. The slope value divided by 4 times the vehicle length is the yaw rate.

[0070] In the section "After filtering out vehicle speed using a formula, only the velocity component caused by angular velocity remains. Its value is linearly related to the sine of the horizontal observation angle. The slope and slant distance of the straight line are obtained using linear regression or RANSAC methods. The slope value divided by 4 times the vehicle length is the yaw rate," after filtering out the vehicle speed in this frame, the radial relative velocity of the point cloud is:

[0071] .

[0072] Specifically, after filtering out the vehicle speed in this frame, the radial relative velocity of the point cloud is:

[0073]

[0074] Since static point clouds are much more numerous than static point clouds in typical scenarios, the slope of the point cloud caused by the turning angular velocity can be obtained through linear regression or the RANSAC method. Figure 2 As shown, when the vehicle's straight-line speed is not filtered out, most of the stationary point cloud will have a trigonometric function relationship with the sine of the horizontal angle. After filtering out the vehicle's straight-line speed, the speed of all stationary point clouds in the straight-line scenario tends to 0 (at this point, the absolute speed has been obtained). In the turning scenario, the stationary point cloud will have a linear function relationship with the sine of the horizontal angle due to the influence of angular velocity. After filtering out the angular velocity using RANSAC or linear regression methods, the vehicle's straight-line speed and turning speed can be estimated.

[0075] S6 obtains the true absolute velocity of the point cloud by filtering out the vehicle's angular velocity from the acquired slope and slant distance.

[0076] Example:

[0077] In a vehicle turning scenario, assuming point cloud data with radial distance, radial relative velocity, and horizontal / pitch angle has been acquired, as shown in Figure 3 (original point cloud velocity diagram. a: 2D point cloud distribution from a top view; b: Relationship between point cloud velocity and the sine of the horizontal angle). In a, the original point cloud is the original point cloud distribution in a 3D Cartesian coordinate system. The vertical point clouds on the left and right sides are the road edges, and vehicles or pedestrians are inside the road edges, presenting a clustered point cloud. Point clouds with an absolute radial relative velocity greater than 0.5 m / s are represented in red. At this time, because the vehicle is turning, the point cloud almost all show a moving state (relative velocity does not represent the actual movement state of the target). In b, the horizontal axis is converted to the sine of the horizontal angle. At this time, the relative velocity of the static point cloud (absolute velocity is 0) will show a trigonometric function shape, while a small part of the dynamic point cloud will not follow this rule.

[0078] Using the formula:

[0079] Estimate the vehicle's straight-line speed. Since the vehicle is turning, the straight-line speed and turning speed are included. After estimating the vehicle speed vcar, filter out the influence of vehicle speed in the point cloud according to the following formula:

[0080]

[0081] get The vehicle's straight-line speed has been filtered out; the residual turning speed causes speed anomalies at large horizontal observation angles, such as... Figure 4 (A schematic diagram of the point cloud after filtering out the speed of straight-moving vehicles. a: 2D point cloud distribution when viewed from above; b: Relationship between point cloud velocity and the sine of the horizontal angle)

[0082] from Figure 4As can be seen, the point cloud exhibits a linear relationship on the sine-relative velocity graph. Linear regression or the RANSAC method can filter out the linear velocity values ​​caused by the vehicle's steering. The resulting slope k includes the angular velocity, and the specific formula is as follows:

[0083]

[0084] ω represents the vehicle's steering angular velocity, and d represents the length from the rear axle center of the vehicle to the radar. Since the vehicle's straight-line speed has been filtered out, the slant distance is approximately 0. The final result is as follows: Figure 5 (Diagram showing the absolute velocity of point clouds. a: 2D point cloud distribution from a top-down view; b: Relationship between point cloud velocity and the sine of the horizontal angle)

[0085] Depend on Figure 5 It can be seen that when the point cloud consists of absolute velocities, the roadside points are static points with a velocity of 0, and the vehicle has a velocity. Only such point clouds can become the input data for the subsequent autonomous driving perception algorithm module.

[0086] The above-disclosed embodiments are merely preferred embodiments of the millimeter-wave radar point cloud velocity decoupling method applicable to turning scenarios of the present invention. Of course, they should not be construed as limiting the scope of the present invention. Those skilled in the art can understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A millimeter-wave radar point cloud velocity decoupling method suitable for turning scenarios, characterized in that, Includes the following steps: The radar echoes are processed to obtain point clouds with velocity, range, and angle information; According to the formula, the radial relative velocity of the point cloud is calculated based on the pitch angle and the horizontal angle; The vehicle speed is obtained by taking the average value of the points near the 0° angle. The obtained vehicle speed is compared with the previous two frames. If the error is greater than ε, the estimated value of the current frame is discarded and the result of the previous frame is used. If there are no previous two frames, the vehicle speed of 10 frames is calculated first, outliers are removed and then the average value is taken as the initial vehicle speed of the previous two frames. By filtering out vehicle speed using a formula, only the velocity component caused by angular velocity remains. Its value is linearly related to the sine of the horizontal observation angle. The slope and slant distance of the straight line are obtained by using linear regression or RANSAC method. The slope value divided by 4 times the vehicle length is the yaw rate. The true absolute velocity of the point cloud is obtained by filtering out the vehicle's angular velocity from the acquired slope and slant distance.

2. The millimeter-wave radar point cloud velocity decoupling method applicable to turning scenarios as described in claim 1, characterized in that, The process of "processing radar echoes to obtain point clouds with velocity, range, and angle information" includes the following steps: The 4D millimeter-wave radar transmits a linear frequency modulated (FMCW) signal. The electromagnetic wave is reflected back to the radar receiver after being reflected by the actual object. After mixing, an intermediate frequency (IF) signal is obtained in the fast time dimension. The radial distance of the target is obtained by performing FFT processing on the mixed signal along the range direction. The Doppler frequency is obtained by performing an FFT in the slow time dimension, and then the radial relative velocity of the target is obtained. The angle is calculated by using the digital beamforming (DBF) method to obtain the distance, velocity and angle information of the point cloud.

3. The millimeter-wave radar point cloud velocity decoupling method applicable to turning scenarios as described in claim 2, characterized in that, In the section "Calculating the radial relative velocity of a point cloud based on pitch and horizontal angles," the formula is: ； Where f d f is the Doppler frequency; c is the speed of light; f c This is the carrier frequency for the transmitted signal.

4. The millimeter-wave radar point cloud velocity decoupling method for turning scenarios as described in claim 3, characterized in that, In the section "taking the average of point cloud data near 0° angle to obtain the vehicle speed", the radial relative velocity of the point cloud acquired by the millimeter-wave radar placed at the front of the vehicle can be expressed as follows: ； Where vcar is the vehicle's straight-line velocity along the 0° normal, vm is the velocity component of the moving target along the 0° normal, ω is the vehicle's steering angular velocity, d is the length from the rear axle center of the vehicle to the radar, θ is the horizontal observation angle, and φ is the pitch observation angle; the Doppler velocity measurement result of the stationary point cloud at the 0° normal position is the vehicle's speed. Therefore, it can be obtained by taking the point cloud in the range of [-5°, 5°], removing the maximum and minimum outliers, and then taking the average, i.e.: ； Where vk is the radial relative velocity of the point cloud within the interval, and n is the number of point clouds within the interval, the vehicle speed can be obtained through the above formula.

5. The millimeter-wave radar point cloud velocity decoupling method for turning scenarios as described in claim 4, characterized in that, In the section "After filtering out vehicle speed using a formula, only the velocity component caused by angular velocity remains. Its value is linearly related to the sine of the horizontal observation angle. The slope and slant distance of the straight line are obtained using linear regression or RANSAC methods. The slope value divided by 4 times the vehicle length is the yaw rate," after filtering out the vehicle speed in this frame, the radial relative velocity of the point cloud is: 。

Images