Millimeter wave radar point cloud real-time processing method and system

Abstract

The invention belongs to the technical field of radars, and relates to a millimeter wave radar point cloud real-time processing method and system. The method comprises the following steps: the millimeter wave radar transmits linear frequency modulation signals of a plurality of pulses; carrying out two-dimensional FFT processing; target detection; obtaining point cloud data according to the target distance, speed and angle information; performing clustering processing to obtain a central point of each cluster; constructing a bounding box; extracting features; calculating a cluster of which the second identification vector is closest to the first identification vector as a point cloud of which the second frame image is matched with the first frame image; performing image registration on the successfully matched point clouds in the first frame image and the second frame image by using a point cloud main direction attachment method; target identification; and filtering the identified targets to obtain the motion trails of the targets. According to the method, the calculation amount of a filtering algorithm is greatly reduced, different targets are subjected to multi-frame point cloud matching through the identification vectors and then are used for classified calculation, and the point cloud recognition precision and the classification accuracy are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and in particular, relates to a millimeter wave radar point cloud real-time processing method and system. Background Art

[0002] The core of millimeter wave imaging radar is high resolution, which means it can distinguish neighboring objects, outline the general outline of the target, and accurately classify and identify the target. It can be used for distance measurement, speed measurement, angle measurement and height measurement. High-density point cloud and high-stability point cloud are the basis for radar point cloud recognition. For the same target, we hope that the number of point clouds detected by the radar is as large as possible, and it is best to be able to describe the outline of the target, which is good for target recognition. However, in fact, radar always has many problems, such as point cloud flickering, point cloud number changes with target distance and point cloud RCS changes, so the target point cloud obtained is unstable.

[0003] The existing method of processing point clouds is to process all the point clouds in a single frame together, and to perform denoising, filtering and target recognition on a large amount of point cloud data at the same time. Therefore, it requires a lot of computing resources, high real-time processing hardware cost, large amount of point cloud data, large amount of real-time processing calculation, poor real-time performance, and high algorithm complexity, which greatly increases hardware costs. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a millimeter wave radar point cloud real-time processing method and system.

[0005] In a first aspect, the present invention provides a method for real-time processing of millimeter wave radar point clouds, comprising: Millimeter wave radar transmits a linear frequency modulation signal with multiple pulses; Perform two-dimensional FFT processing on the echo; Target detection is performed through the CFAR algorithm to obtain target distance and speed parameters, and target angle information is obtained by estimating the target angle between multiple array elements; According to the target distance, speed and angle information, point cloud data is obtained; The clustering threshold is adjusted according to the point cloud density, and the point cloud data is clustered to divide the point cloud data into noise point clouds and several clustered point clouds, and the center point of each cluster is obtained; Construct a bounding box, in which the length, width, and height of the bounding box are defined; Perform feature extraction on the point cloud of the same cluster, including: calculating the centroid of the target point, the average speed of all points in the point cloud of the same cluster, and the density of all points in the point cloud of the same cluster; traversing all clusters in the first frame image, extracting the size information and statistical information of all point clouds, forming a first identification vector, and normalizing the first identification vector; traversing all clusters in the second frame image, extracting the size information and statistical information of all point clouds, forming a second identification vector, and normalizing the second identification vector; Calculate the cluster whose second identification vector of each cluster in the second frame image is closest to the first identification vector of the cluster in the first frame image as the point cloud matching the second frame image with the first frame image, until all point clouds matching the first frame image in the second frame image are obtained; For the successfully matched point clouds in the first frame image and the second frame image, the image registration is performed using the point cloud main direction pasting method; Using the point cloud set to identify the target; The identified targets are filtered to obtain the motion trajectory of each target.

[0006] In a second aspect, the present invention provides a millimeter wave radar point cloud real-time processing system including a transmitting unit, a signal processing unit, a target detection unit, a first data processing unit, a second data processing unit, a construction unit, a feature extraction unit, a third data processing unit, an image registration unit, a recognition unit and an output unit; A transmitting unit, used for transmitting a linear frequency modulation signal of multiple pulses of the millimeter wave radar; A signal processing unit, used for performing two-dimensional FFT processing on the echo; The target detection unit is used to detect the target through the CFAR (Constant False Alarm Rate) algorithm, obtain the target distance and speed parameters, and estimate the target angle between multiple array elements to obtain the target angle information; A first data processing unit, used to obtain point cloud data according to target distance, speed and angle information; The second data processing unit is used to adjust the clustering threshold according to the point cloud density, perform clustering processing on the point cloud data, divide the point cloud data into a noise point cloud and a plurality of clustered point clouds, and obtain the center point of each cluster; A construction unit, used to construct a bounding box, in which the length, width and height of the bounding box are defined; The feature extraction unit is used to extract features from the point cloud of the same cluster, including: calculating the centroid of the target point, the average speed of all points in the point cloud of the same cluster, and the density of all points in the point cloud of the same cluster; traversing all clusters in the first frame image, extracting the size information and statistical information of all point clouds, forming a first identification vector, and normalizing the first identification vector; traversing all clusters in the second frame image, extracting the size information and statistical information of all point clouds, forming a second identification vector, and normalizing the second identification vector; A third data processing unit is used to calculate the cluster whose second identification vector of each cluster in the second frame image is closest to the first identification vector of the cluster in the first frame image, as the point cloud matching the second frame image with the first frame image, until all point clouds matching the first frame image in the second frame image are obtained; An image registration unit, for performing image registration on successfully matched point clouds in the first frame image and the second frame image by using a point cloud main direction pasting method; A recognition unit, which uses the point cloud set to perform target recognition on the target; The output unit is used to filter the identified targets to obtain the motion trajectory of each target.

[0007] Based on the above technical solution, the present invention can also be improved as follows.

[0008] Furthermore, the clustering threshold is adjusted according to the point cloud density, and the point cloud data is clustered, including: Calculate the two points with the closest Euclidean distance to form a data pair; Merge data pairs containing the same data into data groups; Calculate the average value of all the point coordinates in the data group to get the center point of the data group; Calculate the point farthest from the center point between any two data groups to obtain the farthest distance between the two data groups; If the distance between two points in two data groups is less than the set threshold, the two data groups are merged until no data groups are merged.

[0009] Further, construct a bounding box, and define the length, width and height of the bounding box in the bounding box, including: assuming that the length direction of the bounding box is Axis, width direction Axis, height direction is axis, the length of the bounding box is , the width is , the height is , the maximum length of the bounding box is , the minimum length of the bounding box is , the maximum width of the bounding box is , the minimum width of the bounding box is The maximum height of the bounding box is The minimum height of the bounding box is ,but: ; ; .

[0010] Further, suppose is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the number of all points in the point cloud of the same cluster, the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point is defined as an array of the average values ​​of each dimension, let the array be , expressed as ,and: ; ; ; The average velocity of all points in the point cloud of the same cluster is , is the velocity of the ith point, then: ; The density of all points in the point cloud of the same cluster is the ratio of the number of all points in the point cloud of the same cluster to the volume of the cluster. Let the density of all points in the point cloud of the same cluster be , the cluster volume is ,but: .

[0011] Furthermore, let the length of the bounding box be , the width is , the height is , the identification vector of the i-th cluster point cloud is , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the average speed of all points in the point cloud of the same cluster is , the density of all points in the point cloud of the same cluster is ,but: .

[0012] Further, suppose represents the identification vector of the i-th type of point cloud, represents the identification vector of the j-th type of point cloud, represents the vector 1 norm, then: .

[0013] Furthermore, the main direction of the point cloud is used for image registration, including: calculating the feature vectors of all points in the point cloud to obtain the main direction of the point cloud in space and two secondary directions perpendicular to the main direction, and establishing a reference coordinate system with the center of gravity of the point cloud as the origin and the main direction and secondary directions of the point cloud as coordinate axes; the point clouds of different frames are fused according to the reference coordinate system to obtain a set of point clouds of the target at different times as the image registration result.

[0014] The beneficial effects of the present invention are as follows: first, the present invention clusters a large number of point clouds into point targets, which greatly reduces the amount of calculation of the filtering algorithm and enables the system to process in real time; secondly, the identification vectors of different targets are matched in multiple frames of point clouds for classification calculation, which greatly improves the precision of point cloud recognition and classification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a millimeter wave radar point cloud real-time processing method provided in Example 1 of the present invention; Figure 2 A schematic diagram of a millimeter wave radar point cloud real-time processing system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0017] Example 1 As an example, Figure 1 As shown, in order to solve the above technical problems, this embodiment provides a millimeter wave radar point cloud real-time processing method, including: Millimeter wave radar transmits a linear frequency modulation signal with multiple pulses; Perform two-dimensional FFT processing on the echo; Target detection is performed through the CFAR algorithm to obtain target distance and speed parameters, and target angle information is obtained by estimating the target angle between multiple array elements; According to the target distance, speed and angle information, point cloud data is obtained; The clustering threshold is adjusted according to the point cloud density, and the point cloud data is clustered to divide the point cloud data into noise point clouds and several clustered point clouds, and the center point of each cluster is obtained; Construct a bounding box, in which the length, width, and height of the bounding box are defined; Perform feature extraction on the point cloud of the same cluster, including: calculating the centroid of the target point, the average speed of all points in the point cloud of the same cluster, and the density of all points in the point cloud of the same cluster; traversing all clusters in the first frame image, extracting the size information and statistical information of all point clouds, forming a first identification vector, and normalizing the first identification vector; traversing all clusters in the second frame image, extracting the size information and statistical information of all point clouds, forming a second identification vector, and normalizing the second identification vector; Calculate the cluster whose second identification vector of each cluster in the second frame image is closest to the first identification vector of the cluster in the first frame image as the point cloud matching the second frame image with the first frame image, until all point clouds matching the first frame image in the second frame image are obtained; For the successfully matched point clouds in the first frame image and the second frame image, the image registration is performed using the point cloud main direction pasting method; Using the point cloud set to identify the target; The identified targets are filtered to obtain the motion trajectory of each target.

[0018] Optionally, the clustering threshold is adjusted according to the point cloud density to perform clustering processing on the point cloud data, including: Calculate the two points with the closest Euclidean distance to form a data pair; Merge data pairs containing the same data into data groups; Calculate the average value of all the point coordinates in the data group to get the center point of the data group; Calculate the point farthest from the center point between any two data groups to obtain the farthest distance between the two data groups; If the distance between two points in two data groups is less than the set threshold, the two data groups are merged until no data groups are merged.

[0019] The clustering threshold is automatically adjusted according to the point cloud density, so that the present invention is suitable for multi-target clustering in complex scenes. After clustering, the point cloud data is divided into noise points and point clouds of different categories, and the center points of different categories can be obtained.

[0020] The present invention automatically adjusts the clustering threshold according to the point cloud density, performs preliminary preprocessing on the original point cloud, integrates the original point clouds of different targets into one point, and subsequently performs filtering processing on the point cloud. Since the number of processed targets is greatly reduced, a large amount of computing resources is saved. Subsequently, by integrating multiple frames of point clouds of a target into one trajectory, the point cloud information of the target is enriched, and the classification accuracy is effectively improved.

[0021] Optionally, construct a bounding box and define the length, width, and height of the bounding box in the bounding box, including: assuming that the length direction of the bounding box is Axis, width direction Axis, height direction is axis, the length of the bounding box is , the width is , the height is , the maximum length of the bounding box is , the minimum length of the bounding box is , the maximum width of the bounding box is , the minimum width of the bounding box is The maximum height of the bounding box is The minimum height of the bounding box is ,but: ; ; .

[0022] Optional, set is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the number of all points in the point cloud of the same cluster, the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point is defined as an array of the average values ​​of each dimension, let the array be , expressed as ,and: ; ; ; The average velocity of all points in the point cloud of the same cluster is , is the velocity of the ith point, then: ; The density of all points in the point cloud of the same cluster is the ratio of the number of all points in the point cloud of the same cluster to the volume of the cluster. Let the density of all points in the point cloud of the same cluster be , the cluster volume is ,but: .

[0023] Optionally, let the length of the bounding box be , the width is , the height is , the identification vector of the i-th cluster point cloud is , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the average speed of all points in the point cloud of the same cluster is , the density of all points in the point cloud of the same cluster is ,but: .

[0024] Optional, set represents the identification vector of the i-th type of point cloud, represents the identification vector of the j-th type of point cloud, represents the vector 1 norm, then: .

[0025] Optionally, the main direction of the point cloud is used for image registration, including: calculating the feature vectors of all points in the point cloud to obtain the main direction of the point cloud in space and two secondary directions perpendicular to the main direction, establishing a reference coordinate system with the center of gravity of the point cloud as the origin and the main direction and secondary directions of the point cloud as coordinate axes; fusing point clouds of different frames according to the reference coordinate system to obtain a set of point clouds of the target at different times as the image registration result.

[0026] For the point clouds of the two frames of images that have been successfully matched, the main direction of the point cloud is used to achieve automatic registration, which is convenient, fast and accurate. Since the point cloud of the target is accumulated over multiple frames, the number and accuracy of the point cloud are greatly improved, so the recognition accuracy can also be greatly improved.

[0027] The identified target center point is processed by Kalman filtering to obtain the motion trajectory and motion state of different targets. The motion state includes position and speed. At the same time, some clutter signals are filtered out to obtain a more accurate and true target motion trajectory. The point cloud corresponding to the cluster center matching this motion trajectory is associated with this trajectory.

[0028] In complex scenarios, the number of multi-target point clouds acquired by the radar is very large, and the direct filtering and classification algorithm calculation is very computationally intensive. The present invention firstly clusters a large number of point clouds into point targets, which greatly reduces the calculation amount of the filtering algorithm and enables the system to process in real time. Secondly, the identification vectors of different targets are matched through multiple frames of point clouds for classification calculation, which greatly improves the precision of point cloud recognition and classification accuracy.

[0029] Example 2 Based on the same principle as the method shown in Example 1 of the present invention, as shown in the attached Figure 2 As shown, an embodiment of the present invention further provides a millimeter wave radar point cloud real-time processing system, including a transmitting unit, a signal processing unit, a target detection unit, a first data processing unit, a second data processing unit, a construction unit, a feature extraction unit, a third data processing unit, an image registration unit, a recognition unit and an output unit; A transmitting unit, used for transmitting a linear frequency modulation signal of multiple pulses of the millimeter wave radar; A signal processing unit, used for performing two-dimensional FFT processing on the echo; The target detection unit is used to detect the target through the CFAR algorithm, obtain the target distance and speed parameters, and estimate the target angle between multiple array elements to obtain the target angle information; A first data processing unit, used to obtain point cloud data according to target distance, speed and angle information; The second data processing unit is used to adjust the clustering threshold according to the point cloud density, perform clustering processing on the point cloud data, divide the point cloud data into a noise point cloud and a plurality of clustered point clouds, and obtain the center point of each cluster; A construction unit, used to construct a bounding box, in which the length, width and height of the bounding box are defined; The feature extraction unit is used to extract features from the point cloud of the same cluster, including: calculating the centroid of the target point, the average speed of all points in the point cloud of the same cluster, and the density of all points in the point cloud of the same cluster; traversing all clusters in the first frame image, extracting the size information and statistical information of all point clouds, forming a first identification vector, and normalizing the first identification vector; traversing all clusters in the second frame image, extracting the size information and statistical information of all point clouds, forming a second identification vector, and normalizing the second identification vector; A third data processing unit is used to calculate the cluster whose second identification vector of each cluster in the second frame image is closest to the first identification vector of the cluster in the first frame image, as the point cloud matching the second frame image with the first frame image, until all point clouds matching the first frame image in the second frame image are obtained; An image registration unit, for performing image registration on successfully matched point clouds in the first frame image and the second frame image by using a point cloud main direction pasting method; A recognition unit, which uses the point cloud set to perform target recognition on the target; The output unit is used to filter the identified targets to obtain the motion trajectory of each target.

[0030] Optionally, the clustering threshold is adjusted according to the point cloud density to perform clustering processing on the point cloud data, including: Calculate the two points with the closest Euclidean distance to form a data pair; Merge data pairs containing the same data into data groups; Calculate the average value of all the point coordinates in the data group to get the center point of the data group; Calculate the point farthest from the center point between any two data groups to obtain the farthest distance between the two data groups; If the distance between two points in two data groups is less than the set threshold, the two data groups are merged until no data groups are merged.

[0031] Optionally, construct a bounding box and define the length, width, and height of the bounding box in the bounding box, including: assuming that the length direction of the bounding box is Axis, width direction Axis, height direction is axis, the length of the bounding box is , the width is , the height is , the maximum length of the bounding box is , the minimum length of the bounding box is , the maximum width of the bounding box is , the minimum width of the bounding box is The maximum height of the bounding box is The minimum height of the bounding box is ,but: ; ; .

[0032] Optional, set is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the number of all points in the point cloud of the same cluster, the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point is defined as an array of the average values ​​of each dimension, let the array be , expressed as ,and: ; ; ; The average velocity of all points in the point cloud of the same cluster is , is the velocity of the ith point, then: ; The density of all points in the point cloud of the same cluster is the ratio of the number of all points in the point cloud of the same cluster to the volume of the cluster. Let the density of all points in the point cloud of the same cluster be , the cluster volume is ,but: .

[0033] Optionally, let the length of the bounding box be , the width is , the height is , the identification vector of the i-th cluster point cloud is , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the average speed of all points in the point cloud of the same cluster is , the density of all points in the point cloud of the same cluster is ,but: .

[0034] Optional, set represents the identification vector of the i-th type of point cloud, represents the identification vector of the j-th type of point cloud, represents the vector 1 norm, then: .

[0035] Optionally, the main direction of the point cloud is used for image registration, including: calculating the feature vectors of all points in the point cloud to obtain the main direction of the point cloud in space and two secondary directions perpendicular to the main direction, establishing a reference coordinate system with the center of gravity of the point cloud as the origin and the main direction and secondary directions of the point cloud as coordinate axes; fusing point clouds of different frames according to the reference coordinate system to obtain a set of point clouds of the target at different times as the image registration result.

[0036] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A millimeter wave radar point cloud real-time processing method, characterized in that: include: Millimeter wave radar transmits a linear frequency modulation signal with multiple pulses; Perform two-dimensional FFT processing on the echo; Target detection is performed through the CFAR algorithm to obtain target distance and speed parameters, and target angle information is obtained by estimating the target angle between multiple array elements; According to the target distance, speed and angle information, point cloud data is obtained; The clustering threshold is adjusted according to the point cloud density, and the point cloud data is clustered to divide the point cloud data into noise point clouds and several clustered point clouds, and the center point of each cluster is obtained; Construct a bounding box, in which the length, width, and height of the bounding box are defined; Perform feature extraction on the point cloud of the same cluster, including: calculating the centroid of the target point, the average speed of all points in the point cloud of the same cluster, and the density of all points in the point cloud of the same cluster; traversing all clusters in the first frame image, extracting the size information and statistical information of all point clouds, forming a first identification vector, and normalizing the first identification vector; traversing all clusters in the second frame image, extracting the size information and statistical information of all point clouds, forming a second identification vector, and normalizing the second identification vector; Calculate the cluster whose second identification vector of each cluster in the second frame image is closest to the first identification vector of the cluster in the first frame image as the point cloud matching the second frame image with the first frame image, until all point clouds matching the first frame image in the second frame image are obtained; For the successfully matched point clouds in the first frame image and the second frame image, the image registration is performed using the point cloud main direction pasting method; Using the point cloud set to identify the target; The identified targets are filtered to obtain the motion trajectory of each target.

2. According to claim 1, a millimeter wave radar point cloud real-time processing method is characterized in that: The clustering threshold is adjusted according to the point cloud density, and the point cloud data is clustered, including: Calculate the two points with the closest Euclidean distance to form a data pair; Merge data pairs containing the same data into data groups; Calculate the average value of all the point coordinates in the data group to get the center point of the data group; Calculate the point farthest from the center point between any two data groups to obtain the farthest distance between the two data groups; If the distance between two points in two data groups is less than the set threshold, the two data groups are merged until no data groups are merged.

3. According to claim 1, a millimeter wave radar point cloud real-time processing method is characterized in that: Construct a bounding box and define the length, width, and height of the bounding box in the bounding box, including: assuming that the length direction of the bounding box is Axis, width direction Axis, height direction is axis, the length of the bounding box is , the width is , the height is , the maximum length of the bounding box is , the minimum length of the bounding box is , the maximum width of the bounding box is , the minimum width of the bounding box is The maximum height of the bounding box is The minimum height of the bounding box is ,but: ； ； 。 4. According to claim 1, a millimeter wave radar point cloud real-time processing method is characterized in that: set up is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the i-th point Axis coordinates, is the number of all points in the point cloud of the same cluster, the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point is defined as an array of the average values ​​of each dimension, let the array be , expressed as ,and: ； ； ； The average velocity of all points in the point cloud of the same cluster is , is the velocity of the ith point, then: ； The density of all points in the point cloud of the same cluster is the ratio of the number of all points in the point cloud of the same cluster to the volume of the cluster. Let the density of all points in the point cloud of the same cluster be , the cluster volume is ,but: 。 5. According to the method for real-time processing of millimeter wave radar point cloud in claim 1, it is characterized in that The length of the bounding box is , the width is , the height is , the identification vector of the i-th cluster point cloud is , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the centroid of the target point The axis coordinates are , the average speed of all points in the point cloud of the same cluster is , the density of all points in the point cloud of the same cluster is ,but: 。 6. The method for real-time processing of millimeter wave radar point cloud according to claim 1, characterized in that: set up Represents the identification vector of the i-th type of point cloud, represents the identification vector of the j-th type of point cloud, represents the vector 1 norm, then: 。 7. The method for real-time processing of millimeter wave radar point cloud according to claim 1, characterized in that: Image registration is performed using the point cloud main direction pasting method, including: calculating the eigenvectors of all points in the point cloud to obtain the main direction of the point cloud in space and two secondary directions perpendicular to the main direction, establishing a reference coordinate system with the point cloud center of gravity as the origin and the main direction and secondary directions of the point cloud as coordinate axes; fusing point clouds from different frames according to the reference coordinate system to obtain a set of point clouds of the target at different times as the image registration result.

8. A millimeter wave radar point cloud real-time processing system, characterized in that: It includes a transmitting unit, a signal processing unit, a target detection unit, a first data processing unit, a second data processing unit, a construction unit, a feature extraction unit, a third data processing unit, an image registration unit, a recognition unit and an output unit; A transmitting unit, used for transmitting a linear frequency modulation signal of multiple pulses of the millimeter wave radar; A signal processing unit, used for performing two-dimensional FFT processing on the echo; The target detection unit is used to detect the target through the CFAR algorithm, obtain the target distance and speed parameters, and estimate the target angle between multiple array elements to obtain the target angle information; A first data processing unit, used to obtain point cloud data according to target distance, speed and angle information; The second data processing unit is used to adjust the clustering threshold according to the point cloud density, perform clustering processing on the point cloud data, divide the point cloud data into a noise point cloud and a plurality of clustered point clouds, and obtain the center point of each cluster; A construction unit, used to construct a bounding box, in which the length, width and height of the bounding box are defined; The feature extraction unit is used to extract features from the point cloud of the same cluster, including: calculating the centroid of the target point, the average speed of all points in the point cloud of the same cluster, and the density of all points in the point cloud of the same cluster; traversing all clusters in the first frame image, extracting the size information and statistical information of all point clouds, forming a first identification vector, and normalizing the first identification vector; traversing all clusters in the second frame image, extracting the size information and statistical information of all point clouds, forming a second identification vector, and normalizing the second identification vector; A third data processing unit is used to calculate the cluster whose second identification vector of each cluster in the second frame image is closest to the first identification vector of the cluster in the first frame image, as the point cloud matching the second frame image with the first frame image, until all point clouds matching the first frame image in the second frame image are obtained; An image registration unit, for performing image registration on successfully matched point clouds in the first frame image and the second frame image by using a point cloud main direction pasting method; A recognition unit, which uses the point cloud set to perform target recognition on the target; The output unit is used to filter the identified targets to obtain the motion trajectory of each target.

Images