A linear track association method for traffic radar

Abstract

This invention proposes a linear trajectory association method for traffic radar, overcoming the limitation of high computational complexity in conventional trajectory association using nonlinear Kalman filtering. Its main steps include: S1, associating trajectory m with point cloud data n, where the association threshold length corresponds to the extreme value of the second derivative of the k-distance function of point cloud data n; S2, for any associated trajectory m, calculating the measurement value of its corresponding point cloud data n, and calculating the measurement matrix Z of the current trajectory m to characterize the current measurement value; S3, constructing the prediction matrix A, measurement covariance matrix R, and prediction covariance matrix Q for the corresponding traffic vehicles; S4, calculating the linear Kalman filter, the estimation matrix X of the previous trajectory m, and the next estimation matrix X_, where X is estimated from the measurement value of trajectory m and the previous prediction value, and X_ is estimated from the current prediction value; S5, repeating the above steps until all trajectory associations are completed.

Description

Technical Field

[0001] This invention relates to the field of radar data signal processing technology, and in particular to a linear track association method for traffic radar. Background Technology

[0002] The trajectory association problem refers to the process of estimating the true values ​​of measurements using measurement data from one or more noisy sensors and existing prior knowledge. For traffic radar, the radial distance, horizontal azimuth, Doppler frequency, and signal amplitude of the target are obtained from the radar echo, and the true two-dimensional position, speed, and heading of the target are estimated.

[0003] Track association is a crucial step in traffic radar data processing. Singer and Kanyuch first proposed a weighted track association method, which Bar-Shalom modified with a revised algorithm. Building upon this, Kosaka et al. presented a nearest-neighbor track association method. Bowman described a track association method using maximum likelihood estimation. Youens et al. derived a track association algorithm based on classical allocation from operations research. Currently, commonly used track association methods include: Kalman filter weighted algorithm, optimal distributed track association algorithm, and fuzzy track fusion.

[0004] Kalman filtering is a commonly used estimation and prediction method in flight path association. Basic Kalman filtering can only handle linear models, while conventional nonlinear Kalman filtering for flight path association problems involves significant computational overhead. Therefore, we propose a linear flight path association method for traffic radar that can approximate a linear Kalman filter. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the defects of the existing technology. This invention proposes a linear track association method for traffic radar that can use repeated point clouds to approximate linear Kalman filtering.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a linear track association method for traffic radar, comprising the following steps:

[0007] S1. Correlate the track m and the point cloud data n. The correlation threshold length corresponds to the extreme value of the second derivative of the k-distance function of the point cloud data n.

[0008] S2. For any associated track m, calculate the measurement value of its corresponding point cloud number n, and calculate the measurement matrix Z of the current track m from the measurement value to characterize the current measurement value;

[0009] S3. Construct the prediction matrix A, measurement covariance matrix R, and prediction covariance matrix Q for the corresponding traffic vehicles. Here, A is a fixed matrix used to predict the vehicle motion state, and R and Q are both used to record the preset measurement covariance of the vehicle state corresponding values. R is used to characterize the error and correspondence of Z, and Q is used to characterize the error and correspondence of A.

[0010] S4. Calculate the linear Kalman filter, the estimation matrix X of the previous track m, and the estimation matrix X_ of the next track m, where X is estimated by the measured value of track m and the previous prediction value, and X_ is estimated by the current prediction value.

[0011] S5. Repeat the above steps until all track associations are complete.

[0012] Furthermore, in step S1, the specific steps for associating the track m and the point cloud data n are as follows:

[0013] For any track m, calculate its cost function C(m,n) with its corresponding point cloud data n. If C(m,n) is less than the association threshold length in S1, the current point cloud data n is considered to be related to track m; otherwise, it is considered to be unrelated. The point cloud data n corresponding to each track m includes 0 or more.

[0014] Furthermore, in step S1, each point cloud data includes four-dimensional information: distance R, azimuth A, Doppler D, and amplitude SNR. The cost function for associating track m and point cloud n is...

[0015]

[0016] Where Rthreshold, Athreshold, and Dthreshold are the associated threshold lengths, t is the index of track m, and p is the index of point cloud data n.

[0017] Further, in step S2, the calculation steps for the measurement value of the current data frame of track m are as follows:

[0018]

[0019]

[0020]

[0021] Where t-meas represents the measured value of the current track t, and Dp(n) is the Doppler frequency.

[0022] Furthermore, in step S3, the calculation steps for the measurement matrix Z of track m are as follows:

[0023] Z=[Rx(t) Ry(t) vx(t) vy(t) ax(t) ay(t)] T

[0024] vx(t) = Rx(t) - Rx(t-1),

[0025] vy(t) = Ry(t) - Ry(t-1),

[0026] ax(t)=Rx(t)-2Rx(t-1)+Rx(t-2),

[0027] ay(t)=Ry(t)-2Ry(t-1)+Ry(t-2)

[0028] in,

[0029]

[0030] Among them, Rx / Ry / Vs / Vy / Ax / Ay are used to represent the motion state of the vehicle.

[0031] Further, in step S4, the calculation steps for the prediction matrix A are as follows:

[0032]

[0033] Where ΔT represents the time interval.

[0034] Further, in step S4, the measurement covariance matrix R and the prediction covariance matrix Q are constructed as follows:

[0035]

[0036]

[0037] Where Rfac is the observation variance and Qfac is the prediction variance.

[0038] Furthermore, in step S4, the calculation steps for the estimated matrix X and the next estimated matrix X_ are as follows:

[0039] X_=AX

[0040] P_=APA T +Q

[0041] K = P_ / (P_ + R)

[0042] X = X_ + K(Z - X_)

[0043] P = (IK) * P_

[0044] Where I is the unit matrix, K is the Kalman gain, P is the estimated covariance matrix, and P_ is the estimated covariance matrix for the next iteration.

[0045] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the linear track association method for traffic radar as described above.

[0046] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is used to implement the steps of the linear track association method for traffic radar described above.

[0047] Compared with existing technologies, the advantages of this invention include: by transforming point cloud information from spherical coordinates to rectangular coordinates, utilizing the high data rate of radar and the characteristic of repeating point clouds, track association is performed through linear Kalman filtering. Furthermore, the computational load of the association process is relatively small due to the linear Kalman filtering calculation; no additional prior information is required, including radar installation height; and parameter adjustment is easier in the rectangular coordinate system. Attached Figure Description

[0048] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0049] Figure 1 The flowchart illustrating the linear trajectory association method for traffic radar point cloud data according to the present invention is shown in the schematic diagram.

[0050] Figure 2 The diagram illustrates the measured results of the linear trajectory association method based on traffic radar point cloud data according to the present invention. Detailed Implementation

[0051] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0052] Overall, in this embodiment, the method for associating linear tracks of traffic radar includes the following steps:

[0053] S1. Correlate the track m and the point cloud data n. The correlation threshold length corresponds to the extreme value of the second derivative of the k-distance function of the point cloud data n.

[0054] S2. For any associated track m, calculate the measurement value of its corresponding point cloud number n, and calculate the measurement matrix Z of the current track m from the measurement value to characterize the current measurement value;

[0055] S3. Construct the prediction matrix A, measurement covariance matrix R, and prediction covariance matrix Q for the corresponding traffic vehicles. Here, A is a fixed matrix used to predict the vehicle motion state, and R and Q are both used to record the preset measurement covariance of the vehicle state corresponding values. R is used to characterize the error and correspondence of Z, and Q is used to characterize the error and correspondence of A.

[0056] S4. Calculate the linear Kalman filter, the estimation matrix X of the previous track m, and the estimation matrix X_ of the next track m, where X is estimated by the measured value of track m and the previous prediction value, and X_ is estimated by the current prediction value.

[0057] S5. Repeat the above steps until all track associations are complete.

[0058] The following is in conjunction with the appendix Figure 1 To further explain the above steps, in step S1, the linear trajectory association method for traffic radar point cloud data specifically includes the following steps:

[0059] For traffic radar point cloud data, including four-dimensional information such as range R, azimuth A, Doppler D, and amplitude SNR:

[0060] (1) Correlate the track m and the point cloud n. The cost function for the correlation is:

[0061]

[0062] The associated threshold lengths Rthreshold, Athreshold, and Dthreshold are determined by the extrema of the second derivative of the k-distance function corresponding to the threshold, where t is the subscript of track m and p is the subscript of point cloud data n.

[0063] (2) For the track m and point cloud n corresponding to the threshold, calculate the measurement value of the current data frame of track m:

[0064]

[0065]

[0066]

[0067] Here, t-meas represents the measured value of the current track t, and the Doppler frequency Dp(n) needs to be de-Doppler blurred.

[0068] (3) The calculation steps for the measurement matrix Z of track m are as follows.

[0069] Z=[Rx(t) Ry(t) vx(t) vy(t) ax(t) ay(t)] T

[0070] vx(t) = Rx(t) - Rx(t-1),

[0071] vy(t) = Ry(t) - Ry(t-1),

[0072] ax(t)=Rx(t)-2Rx(t-1)+Rx(t-2),

[0073] ay(t)=Ry(t)-2Ry(t-1)+Ry(t-2)

[0074] in

[0075]

[0076] Among them, Rx / Ry / Vs / Vy / Ax / Ay are used to represent the motion state of the vehicle.

[0077] (4) Construct the prediction matrix A

[0078]

[0079] Where ΔT represents the time interval.

[0080] (5) Construct the measurement covariance matrix R and the prediction covariance matrix Q.

[0081]

[0082]

[0083] Where Rfac is the observation variance and Qfac is the prediction variance.

[0084] (6) Calculate the linear Kalman filter. The steps for calculating the estimated matrix X and the next estimated matrix X_ are as follows.

[0085] X_=AX

[0086] P_=APA T +Q

[0087] K = P_ / (P_ + R)

[0088] X = X_ + K(Z - X_)

[0089] P = (IK) * P_

[0090] Where I is the unit matrix, K is the Kalman gain, P is the estimated covariance matrix, and P_ is the next estimated covariance matrix.

[0091] (7) Repeat steps 1 to 6 until all track associations are completed.

[0092] The invention will be further described in detail below through engineering examples.

[0093] The radar system used in the engineering example is the LFMCW radar, which operates in the 77GHz band. It has 12 transmitting antennas and 16 receiving antennas, with a transmitting power of about 5dBm and a maximum range of about 300m against vehicles.

[0094] The actual trajectory correlation effect in engineering is as follows Figure 2 As shown. Figure 2 The red text represents point cloud data, and the blue text represents track association results. Figure 2 From about 300 meters onward, oncoming vehicles gradually slowed down and changed lanes to the right. Figure 2 This demonstrates the applicability and effectiveness of the linear trajectory association method for traffic radar point cloud data.

[0095] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A linear track association method for traffic radars, characterized in that, The method comprises the following steps: S1, associating the track m and the point cloud data n, and the association threshold length is corresponding to the extreme value of the second derivative of the k-distance function of the point cloud data n; S2, calculating the measurement value of the corresponding point cloud data n of any associated track m, and calculating the measurement matrix Z of the current track m from the measurement value to represent the current measurement value; S3, constructing the prediction matrix A, the measurement covariance matrix R and the prediction covariance matrix Q of the corresponding traffic vehicle, wherein A is a fixed matrix used for predicting the vehicle motion state, R and Q are used for recording the preset measurement covariance of the vehicle state corresponding value, and R is used to represent the error and the corresponding relationship of Z, and Q is used to represent the error and the corresponding relationship of A; S4, calculating the estimation matrix X of the previous track m and the next estimation matrix X_ of the linear Kalman filter, wherein X is estimated from the measurement value of the track m and the previous prediction value, and X_ is estimated from the current prediction value; S5, repeating the above steps until all track associations are completed.

2. The linear track association method for traffic radar according to claim 1, characterized in that: In the step S1, the specific steps of associating the track m and the point cloud data n are as follows, For any track m, the cost function C(m, n) of the track m and the corresponding point cloud data n is calculated, if C(m, n) is less than the association threshold length in S1, it is considered that the current point cloud data n is related to the track m, otherwise it is considered that it is not related, wherein the point cloud data n corresponding to each track m includes 0 or more.

3. The linear track association method for traffic radar according to claim 2, characterized in that: In the step S1, each point cloud data includes four-dimensional information of distance R, azimuth A, Doppler D and amplitude SNR, The cost function of the track m and the point cloud n is Wherein, Rthreshold, Athreshold, Dthreshold are the threshold length of association, t is the subscript of the track m, and p is the subscript of the point cloud data n.

4. The linear track association method for traffic radar according to claim 1, characterized in that: In the step S2, the measurement value calculation step of the current data frame of the track m is as follows, Wherein, t-meas indicates the measurement value of the current track t, and Dp(n) is the Doppler frequency.

5. The linear track association method for traffic radar according to claim 1, characterized in that: In the step S3, the measurement matrix Z of the track m is calculated as follows, Z = [Rx(t) Ry(t) vx(t) vy(t) ax(t) ay(t)] T vx(t)=Rx(t)-Rx(t-1), vy(t)=Ry(t)-Ry(t-1), ax(t)=Rx(t)-2Rx(t-1)+Rx(t-2), ay(t)=Ry(t)-2Ry(t-1)+Ry(t-2) Wherein, Wherein, Rx / Ry / Vs / Vy / Ax / Ay are used to represent the motion state of the vehicle.

6. The linear track association method for traffic radar according to claim 1, characterized in that: In the step S4, the prediction matrix A is calculated as follows, Wherein, ΔT represents the time interval.

7. The linear track association method for traffic radar according to claim 1, characterized in that: In the step S4, the measurement covariance matrix R and the prediction covariance matrix Q are constructed as follows, Wherein, Rfac is the observation variance, and Qfac is the prediction variance.

8. The linear track association method for traffic radar according to claim 1, characterized in that: In the step S4, the estimation matrix X and the next estimation matrix X_ are calculated as follows, X_=AX P_ = APA T +Q K=P_ / (P_+R) X=X_+K(Z-X_) P=(I-K)*P_ Wherein, I is the unit matrix, K is the Kalman gain, P is the estimation covariance matrix, and P_ is the next estimation covariance matrix.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that The processor, when executing the program, implements the steps of the linear track correlation method for traffic radars as claimed in any of claims 1 to 8. 10.A non-transitory computer-readable storage medium having stored thereon a computer program. The computer program, when executed by a processor, implements the steps of the linear track correlation method for traffic radars as claimed in any of claims 1 to 8.

Images