A high-resolution point cloud imaging radar tracking filtering method for extended targets

Abstract

The application belongs to the technical field of tracking filtering, and discloses a tracking filtering method for extended targets by a high-resolution point cloud imaging radar, wherein the fusion filtering takes the Kalman filtering as an example, and the filtering processing of each sub-section representative point is embodied in steps three to ten; in addition, after the state estimation of each sub-section representative point is completed in step nine, the average of all the estimation results is obtained to be a final output, which is taken as the filtering result of the final representative point of the point cloud target. The tracking filtering method for extended targets by the high-resolution point cloud imaging radar can improve the tracking precision, reduce the tracking error and improve the positioning accuracy compared with the traditional tracking filtering method.

Description

Technical Field

[0001] This invention belongs to the field of tracking and filtering technology, specifically relating to a tracking and filtering method for extended targets by a high-resolution point cloud imaging radar. Background Technology

[0002] The conventional extended target tracking method selects the center point of the target to represent the entire target for tracking filtering. The detailed process is as follows: After creating a new trajectory for the target to be tracked, each time target points are matched and associated, the center point of the target is selected to represent the entire extended target and associate it with the trajectory. After a successful match, the representative point—the center point—is filtered, and the filtering result is output.

[0003] In the context of current advanced radar technology—high-resolution point cloud radar—denser point clouds, i.e., finer contours, can be formed for extended targets. Therefore, the point cloud contour of an extended target can be divided into N segments. For each segment, a representative point is selected for individual correlation matching. After successful correlation matching, the N segments are fused and filtered. This effectively reduces errors in the filtering stage and improves filtering accuracy. The improvement is particularly significant when the representative point detected for extended targets fluctuates greatly. The fusion filtering of the N segments is a key step. This invention uses the Kalman filter algorithm as a basis to improve the filtering process.

[0004] When using the center point as a representative point for tracking and filtering extended targets, the trajectory filtering accuracy fluctuates significantly when encountering large fluctuations in the representative point, leading to increased filtering error and weak robustness. Against a high-resolution point cloud background, the target outline is clearer. Dividing a point cloud target into multiple sub-segments and selecting a representative point for matching each sub-segment before fusion filtering can reduce the detection fluctuations of individual representative points and improve filtering accuracy. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention proposes a high-resolution point cloud imaging radar tracking and filtering method for extended targets, in order to improve tracking accuracy, reduce tracking error, and enhance positioning accuracy.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0007] A high-resolution point cloud imaging radar tracking and filtering method for extended targets includes the following steps:

[0008] Step 1: After radar signal processing, the target point cloud information matrix M is obtained. The point data in matrix M are sorted in ascending order according to the distance term y. i Choose a center point C from (i = 1, 2, ..., N) i(i = 1, 2, ... N) are used as representative points;

[0009] Step 2: Select the center point C i (i = 1, 2, ... N) are matched and associated with existing trajectories;

[0010] Step 3: Create a new trajectory, and assign a sub-segment representative point C. i A central representative point C is obtained by averaging the three-dimensional coordinates of (i = 1, 2, ..., N). cen This point is used as the representative point for the location information of the entire point cloud;

[0011] Step 4: Point matching successful. Use the segment stored after the last successful match within the trajectory to represent point C. lasti (i = 1, 2, ... N) Perform one-step state prediction

[0012] C pre =F(n,n-1)C last

[0013] Among them, C last C lasti A matrix consisting of (i = 1, 2, ..., N) is called C. pre The matrix after state prediction has the same mathematical structure as the matrix. F(n,n-1) is the state transition matrix, and its mathematical structure is as follows:

[0014]

[0015] Step 5: Use each matched segment to represent point C i (i = 1, 2, ..., N) perform the process of calculating the new information.

[0016] α = CC pre

[0017] Where C is C i A matrix consisting of (i = 1, 2, ..., N);

[0018] Step Six: For each segment representative point C i (i = 1, 2, ... N) Calculate the one-step prediction error autocorrelation matrix

[0019] P(n,n-1)=F(n,n-1)P(n-1)F H (n, n-1)

[0020] +Γ(n,n-1)Q1Γ H (n, n-1)

[0021] Where Q1 is the state noise matrix, which is related to the target acceleration, Γ(n,n-1) is the noise input matrix, and P(n-1) is the state error autocorrelation matrix;

[0022] Step 7: For each segment, represent point C i (i = 1, 2, ..., N) Calculate the autocorrelation matrix of the information process.

[0023] A(n)=C(n)P(n,n-1)C H (n)+Q2(n)

[0024] Where C(n) is the state transition matrix and Q2(n) is the observation noise matrix, which is related to the radar measurement accuracy;

[0025] Step 8: For each segment, represent point C i Calculate the Kalman gain (i = 1, 2, ..., N)

[0026]

[0027] Step 9: For each segment, represent point C i (i = 1, 2, ..., N) perform state estimation

[0028] C out =C pre +K(n)α(n)

[0029] Obtain the state estimate C of each segment representing a point. out Then, perform an average operation on it.

[0030]

[0031] The value C out_cen As the final output representative point, it represents the filtered result of the center position of the point cloud target.

[0032] Step 10: Update the error autocorrelation matrix

[0033] P(n) = [IK(n)C(n)]P(n,n-1)

[0034] Where I is the identity matrix.

[0035] Furthermore, each point cloud in step one has three-dimensional coordinates (x, y, z).

[0036] Furthermore, in step one, the point cloud information matrix M is divided into N equal parts, thus obtaining the sub-point cloud information matrices M1, M2, ..., M N Then, in each sub-point cloud matrix M i Choose a center point C from (i = 1, 2, ..., N) i(i = 1, 2, ... N) are used as representative points.

[0037] Furthermore, in step two, the nearest neighbor association method and the probabilistic data association method are used.

[0038] Furthermore, if a match is not found in step two, a new trajectory is created and the process proceeds to step three; if a match is found, a filtering operation is performed and the process proceeds to step four.

[0039] Furthermore, in step three, this point is also used for subsequent trajectory and detection point association matching, while representing point C of each sub-segment. i (i = 1, 2, ... N) are stored in this trajectory array.

[0040] Compared with similar existing technologies, the present invention has the following technical advantages:

[0041] (1) The present invention can divide the point cloud contour of an extended target into N segments, select a representative point for each sub-segment for separate corresponding association matching, and perform fusion filtering on the N sub-segments after successful association matching.

[0042] (2) Taking Kalman filtering as an example, the fusion filtering of the present invention is reflected in steps three to ten as filtering processing of each segment representative point. In addition, in step nine, after the state estimation of each segment representative point is completed, the average of all estimation results is calculated to obtain a final output, which is the filtering result of the final representative point of a point cloud target.

[0043] (3) This invention proposes a tracking and filtering method for extended targets using a high-resolution point cloud imaging radar. Compared with traditional tracking and filtering methods, this method can improve tracking accuracy, reduce tracking error, and enhance positioning accuracy.

[0044] (4) When this invention uses the center point as a representative point for tracking and filtering extended targets, the trajectory filtering accuracy will fluctuate significantly when encountering large fluctuations in the representative point, and the filtering error will increase. The robustness is weak. Against a high-resolution point cloud background, the target outline is clearer. Dividing a point cloud target into multiple sub-segments and selecting a representative point for matching in each sub-segment before fusion filtering can reduce the detection fluctuation of a single representative point and improve filtering accuracy. Attached Figure Description

[0045] Figure 1 This is a flowchart of the high-resolution point cloud imaging radar's tracking and filtering method for extended targets according to the present invention. Detailed Implementation

[0046] like Figure 1 As shown, a high-resolution point cloud imaging radar tracking and filtering method for extended targets includes the following steps:

[0047] Step 1: After radar signal processing, a target point cloud information matrix M is obtained, where each point cloud has three-dimensional coordinates (x, y, z). The point data in matrix M are sorted in ascending order according to the distance term y, and then divided into N equal parts, thus obtaining the sub-point cloud information matrices M1, M2, ..., M N Then, in each sub-point cloud matrix M i Choose a center point C from (i = 1, 2, ..., N) i (i = 1, 2, ... N) are used as representative points;

[0048] Step 2: Select the center point C i (i = 1, 2, ... N) is matched and associated with existing trajectories. Nearest neighbor association method, probabilistic data association method, etc. can be used. If the match is unsuccessful, a new trajectory is created and the process proceeds to step three. If the match is successful, a filtering operation is performed and the process proceeds to step four.

[0049] Step 3: Create a new trajectory, and assign a sub-segment representative point C. i A central representative point C is obtained by averaging the three-dimensional coordinates of (i = 1, 2, ..., N). cen This point is used as the representative point of the entire point cloud's location information, and is also used for subsequent trajectory and detection point association matching. Simultaneously, each sub-segment's representative point C... i (i = 1, 2, ... N) are stored in this trajectory array;

[0050] Step 4: Point matching successful. Use the segment stored after the last successful match within the trajectory to represent point C. lasti (i = 1, 2, ... N) Perform one-step state prediction

[0051] C pre =F(n,n-1)C last

[0052] Among them, C last C lasti A matrix consisting of (i = 1, 2, ..., N) is called C. pre The matrix after state prediction has the same mathematical structure as the matrix. F(n,n-1) is the state transition matrix, and its mathematical structure is as follows:

[0053]

[0054] Step 5: Use each matched segment to represent point C i (i = 1, 2, ..., N) perform the process of calculating the new information.

[0055] α = CC pre

[0056] Where C is C i A matrix consisting of (i = 1, 2, ..., N);

[0057] Step Six: For each segment representative point C i (i = 1, 2, ... N) Calculate the one-step prediction error autocorrelation matrix

[0058] P(n,n-1)=F(n,n-1)P(n-1)F H (n, n-1)

[0059] +Γ(n,n-1)Q1Γ H (n, n-1)

[0060] Where Q1 is the state noise matrix, which is related to the target acceleration, Γ(n,n-1) is the noise input matrix, and P(n-1) is the state error autocorrelation matrix;

[0061] Step 7: For each segment, represent point C i (i = 1, 2, ..., N) Calculate the autocorrelation matrix of the information process.

[0062] A(n)=C(n)P(n,n-1)C H (n)+Q2(n)

[0063] Where C(n) is the state transition matrix and Q2(n) is the observation noise matrix, which is related to the radar measurement accuracy;

[0064] Step 8: For each segment, represent point C i Calculate the Kalman gain (i = 1, 2, ..., N)

[0065]

[0066] Step 9: For each segment, represent point C i (i = 1, 2, ..., N) perform state estimation

[0067] C out =C pre +K(n)α(n)

[0068] Obtain the state estimate C of each segment representing a point. out Then, perform an average operation on it.

[0069]

[0070] The value C out_cen As the final output representative point, it represents the filtered result of the center position of the point cloud target.

[0071] Step 10: Update the error autocorrelation matrix

[0072] P(n) = [IK(n)C(n)]P(n,n-1)

[0073] Where I is the identity matrix.

[0074] After steps one through nine, the matching filtering of the detection point cloud results for a target is completed.

[0075] When this invention uses the center point as a representative point for tracking and filtering extended targets, the trajectory filtering accuracy fluctuates significantly when encountering large fluctuations in the representative point, leading to increased filtering error and weak robustness. Against a high-resolution point cloud background, the target outline is clearer. Dividing a point cloud target into multiple sub-segments and selecting a representative point for matching each sub-segment before fusion filtering can reduce the detection fluctuations of individual representative points and improve filtering accuracy.

[0076] The above description is merely a specific embodiment of the present invention for better understanding. The scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-resolution point cloud imaging radar method for tracking and filtering extended targets, characterized in that, Includes the following steps: Step 1: After radar signal processing, the target point cloud information matrix is ​​obtained. , matrix The point data in the data is arranged according to the distance item. Sort from smallest to largest, from point cloud matrix M i Select a center point C i As a representative point, i=1,2,...N ; Step 2: Select the center point C i Match and associate with existing trajectories; Step 3: Create a new trajectory and assign representative points to sub-segments. C i A central representative point is obtained by averaging the three-dimensional coordinates. This point is used as the representative point for the location information of the entire point cloud; Step 4: Point matching successful. Use the segment stored after the last successful match within the trajectory to represent the point. Perform state one-step prediction ； in, for The matrix formed The matrix represents the predicted state; both have the same mathematical structure. The state transition matrix has the following mathematical structure: ； Step 5: Use each matched segment to represent a point C i Performing the process of calculating new information ； in, for C i The matrix formed; Step Six: For each segment representative point C i Calculate the one-step prediction error autocorrelation matrix ； in, This is the state noise matrix, which is related to the target acceleration. The noise input matrix, This is the state error autocorrelation matrix; Step 7: For each segment representative point C i Calculate the autocorrelation matrix of the information process. ； in, Here is the state transition matrix. The observation noise matrix is ​​related to the radar measurement accuracy. Step 8: For each segment representative point C i Calculate Kalman gain ； Step 9: For each segment representative point C i Perform state estimation ； Obtain the state estimate of the point represented by each sub-segment. Then, perform an averaging operation on it. ； This value As the final output representative point, it represents the filtered result of the center position of the point cloud target. Step 10: Update the error autocorrelation matrix ； in, It is an identity matrix.

2. The high-resolution point cloud imaging radar tracking and filtering method for extended targets according to claim 1, characterized in that, In step one, each point cloud has three-dimensional coordinates. .

3. The high-resolution point cloud imaging radar tracking and filtering method for extended targets according to claim 1, characterized in that, In step one, the point cloud information matrix is... Divide it into N equal parts to obtain the sub-point cloud information matrix. Then in each sub-point cloud matrix M i Select a center point C i As a representative point.

4. The high-resolution point cloud imaging radar tracking and filtering method for extended targets according to claim 1, characterized in that, Step two employs the nearest neighbor association method and the probabilistic data association method.

5. The high-resolution point cloud imaging radar tracking and filtering method for extended targets according to claim 1, characterized in that, If a match is not found in step two, a new trajectory is created and the process proceeds to step three; if a match is found, a filtering operation is performed and the process proceeds to step four.

6. The high-resolution point cloud imaging radar tracking and filtering method for extended targets according to claim 1, characterized in that, In step three, this point is also used for subsequent trajectory and detection point association matching, and at the same time, it represents the point of each sub-segment. C i It is stored in the trajectory array.

Images