An autonomous vehicle and a target perception method and system thereof

Abstract

The application provides an automatic driving vehicle and a target sensing method and system thereof, and belongs to the technical field of automatic driving vehicle target sensing. First, original point cloud data of a vehicle surrounding environment is acquired and preprocessed; then, the preprocessed point cloud data is clustered to realize target segmentation and generate target polygons; each target polygon is simplified, including short edge simplification, reliable edge simplification and invalid edge simplification; the application maximally retains shape features of target objects; since the retained reliable edges are edges close to a radar, the shape features of the target objects can be more accurately represented, the accuracy of target object description is more guaranteed, compared with the target description by using a convex polygon in the prior art, the application reduces data storage capacity, and also reduces calculation amount in subsequent collision detection or trajectory planning, so that the shape features of the target objects and the data amount operation problem are considered, and accurate and reliable data is provided for automatic driving path planning and control.

Description

TECHNICAL FIELD

[0001] The present application relates to an autonomous vehicle and a target perception method and system thereof, and belongs to the technical field of target perception of autonomous vehicles. BACKGROUND

[0002] With the development of artificial intelligence technology and computer technology, unmanned driving technology is becoming mature. Environmental perception is the eyes of unmanned driving and the basis for the development of autonomous driving, which can realize accurate perception of the environment around the driving vehicle and provide rich information for downstream module system decision-making. Laser radar can capture rich three-dimensional spatial information and obtain features such as the outline of the target, but the data is original and huge in quantity, and how to quickly and accurately extract and detect the target through related technology is a problem. There are mainly two target expression methods in the prior art, one is to use a convex polygon to describe the target, different target point clouds are segmented by clustering the original point clouds, and then convex polygon point detection is performed on the target point clouds to generate the convex polygon of the target; the other is to use a quadrilateral box to describe the target.

[0003] Among them, the convex polygon is used to describe the target, which can accurately describe the shape features of the target object, but the number of edges of the convex polygon is large, and the data storage amount is large, which makes the calculation amount in the subsequent collision detection or trajectory planning of the autonomous vehicle too large; and the quadrilateral is used to describe the target, the data storage amount is small and the subsequent calculation amount is also small, but the description of the target using the quadrilateral is not accurate enough, and the accurate shape features of the target cannot be provided, which will affect the judgment of the subsequent system decision and the safety of the unmanned driving vehicle. SUMMARY

[0004] The purpose of the present application is to provide an autonomous vehicle and a target perception method and system thereof, so as to solve the problem that the prior art cannot balance the shape features of the target object and the data storage amount and the calculation amount.

[0005] The present application provides a target perception method of an autonomous vehicle, which comprises the following steps:

[0006] 1) obtaining the original point cloud data of the vehicle surrounding environment through the radar, and pre-processing the original point cloud data; the pre-processing includes unifying the space coordinate system and time of all original point cloud data;

[0007] 2) clustering the pre-processed point cloud data to realize target segmentation, obtaining target point cloud data, and generating target polygons based on the point cloud data of each target;

[0008] 3) for each target polygon, determining the reliable edge of each target according to the radar coordinates and the position of the polygon vertex, the reliable edge refers to the edge close to the radar;

[0009] 4) simplifying reliable edges of each target polygon: counting the number of reliable edges in the target polygon and the length of each reliable edge, finding the longest edges of a set number of reliable edges; extracting reliable edges according to a set number of intervals in a set order, taking the extracted reliable edges and the longest edges of a set number as the initialization of the simplified edges; verifying the rationality according to the initialization of the simplified edges, taking the initialization of the simplified edges and the remaining edges that do not meet the rationality verification as the simplification result of the reliable edges of the target polygon;

[0010] The rationality verification process is: calculating the deviation degree of the intersection points of two adjacent edges in the initialization of the simplified edges, when the deviation degree of the intersection points of the two adjacent edges is greater than the set deviation threshold, the reliable edge between the two adjacent edges is retained; when the deviation degree of the intersection points of the adjacent edges is less than the set deviation threshold, the reliable edge between the two adjacent edges is deleted.

[0011] The application further provides a target sensing system of an autonomous vehicle, which comprises a processor and a memory, and the processor is used for processing a computer program stored in the memory to realize the target sensing method of the autonomous vehicle.

[0012] The application further provides an autonomous vehicle, which comprises a radar and an inertial navigation system arranged on a vehicle body, and the radar and the inertial navigation system are connected with a controller to realize the target sensing method of the autonomous vehicle.

[0013] The application firstly clusters the preprocessed point cloud data, realizes target segmentation, and generates target polygons, then determines reliable edges according to radar coordinates and vertex coordinates of each polygon, simplifies the target polygon according to the reliable edges of each target, retains a certain number of longest reliable edges, extracts reliable edges according to a certain number of intervals, and further verifies the rationality to determine whether to delete the remaining reliable edges; the application maximally retains the shape features of the target object, since the retained reliable edges are edges close to the radar, the accuracy of these reliable edges is higher, and the shape features of the target object can be more accurately represented, which further ensures the accuracy of the description of the target object, compared with the convex polygon used in the prior art for target description, the application appropriately reduces the target polygon according to the reliable edges, reduces the data storage amount, and also reduces the calculation amount in subsequent collision detection or trajectory planning, which takes into account the shape features of the target object and the data amount operation, and provides accurate and reliable data for autonomous driving path planning and control.

[0014] Further, the target polygon simplification further comprises simplification of invalid edges, counting the number and length of invalid edges, and retaining the longest edges of a set number of invalid edges; wherein the invalid edge refers to an edge far from the radar.

[0015] The application also simplifies invalid edges in the simplification process. Since the invalid edges are far away from the radar, the accuracy of these invalid edges is not high compared to the valid edges, and the accuracy of the description of the target object is also lower than that of the valid edges. Selecting the longest invalid edges of a certain number of invalid edges and directly deleting other invalid edges will not have a particularly large impact on the overall target polygon, further ensuring the accuracy of the description of the target object.

[0016] Further, before the reliable edge simplification, the short edges need to be simplified first. First, the edges with a length less than a set length threshold are regarded as short edges. It is judged whether the deviation degree of the intersection point of the extensions of the adjacent edges on both sides of the short edge meets a set short edge deviation threshold. When the deviation degree is less than the set short edge deviation threshold, it is determined to delete the short edge. When the deviation degree is greater than the set short edge deviation threshold, the short edge is still retained. The deviation degree refers to the proportion of the distance from the intersection point of the extensions of the adjacent edges on both sides of the short edge to the short edge to the distance from the intersection point to the center point of the polygon.

[0017] Through the above process, the short edges in the polygon are simplified. It is judged whether the short edge can be deleted according to the deviation degree of the intersection point of the extensions of the adjacent edges on both sides of the short edge. Under the premise of ensuring the accurate description of the target shape, unnecessary short edges are deleted, and the data storage amount and subsequent calculation amount are reduced.

[0018] Further, the judgment method of the reliable edge is as follows: finding the maximum angle vertex and the minimum angle vertex in the target polygon, taking the line between the maximum angle vertex and the minimum angle vertex as the first vector, taking the lines between each vertex of the target polygon and the minimum angle vertex as the second vector, and performing cross multiplication on the first vector and the second vector. When the cross multiplication result is greater than 0, the corresponding polygon vertex is a reliable vertex. When the cross multiplication result is less than 0, the corresponding polygon vertex is an unreliable vertex. The edges between the reliable vertices are regarded as reliable edges. The maximum angle vertex refers to the point last scanned by the radar in the scanning process of the target polygon, and the minimum angle vertex refers to the point first scanned by the radar in the scanning process of the target polygon.

[0019] The reliable vertices in the target polygon are determined by vector cross multiplication, and the edges formed by the reliable vertices are regarded as reliable edges, which better ensures the accuracy of the reliable edges.

[0020] Further, before polygon simplification, it is judged whether the target polygon and the vehicle body box exist intersection points, if the intersection points exist, the point cloud number of the target polygon in the first area, the second area and the third area is counted, if the point cloud number in any area is less than the set number of simplified edges, the point cloud in the area is merged into the adjacent area, and the new polygon of the target in different areas is formed.

[0021] When the target polygon and the vehicle body exist covering areas, it does not conform to the actual situation, therefore, through the above process, the area is divided, the point cloud number in different areas is counted, when the point cloud number in a certain area cannot meet the set number of simplified edges, it is merged into the adjacent area, and the new polygon of the target in different areas is finally formed, and the accuracy of target description is ensured.

[0022] Further, the point cloud clustering in the step 2) adopts segmented clustering, at least the point cloud is divided into two segments, and each segment is clustered according to the corresponding set clustering threshold; wherein the set clustering threshold is related to the distance of the point cloud from the radar, when the point cloud is closer to the radar, the clustering threshold is set smaller.

[0023] Because the point cloud is different in distance from the vehicle, the characteristics of dense near and sparse far are formed, therefore, the point cloud at different distances is clustered in the above manner, so that the target segmentation is more accurate; at the same time, the closer the distance, the higher the danger degree of the target to the vehicle body, therefore, the clustering accuracy of the point cloud at different distances is different, the closer the distance, the higher the clustering accuracy, and the smaller the corresponding set clustering threshold.

[0024] Further, the set number in the step 4) is [2, 5], the set number is related to the number of reliable edges, when the number of reliable edges is more, the set number is larger; the set interval number in the step 4) is also related to the number of reliable edges, when the number of reliable edges is more, the set interval number is larger.

[0025] Through the above process, the number of initialization simplified edges is determined according to the number of determined reliable edges, so as to adaptively adjust to different situations.

[0026] Further, the deviation degree is that the vertex of the reliable edge between two adjacent edges in the initialization simplified edge is taken as a target point, the distance from the intersection point of the adjacent edge to the target point accounts for the distance from the intersection point to the center point of the polygon. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 It is the specific flow chart of target perception of the automatic driving vehicle of the application;

[0028] Figure 2is a specific flowchart of the polygon simplification object of the present application.

[0029] Figure 3 An example diagram of reliable edges in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The specific embodiments of the present application are further described below in conjunction with the accompanying drawings.

[0031] Method embodiment

[0032] The present application proposes a target perception method for an autonomous vehicle, and the specific flowchart is as shown in Figure 1 First, the original point cloud data of the vehicle's surrounding environment is obtained, and preprocessed; then the point cloud data after preprocessing is clustered to realize target segmentation; when the polygon generated by the segmented target point cloud has an intersection with the vehicle body box, the target point cloud is split according to the divided area, and a new target polygon in the corresponding area is generated; finally, each target polygon is simplified, and the specific simplification flowchart is as shown in Figure 2

[0033] Step 1. Obtain original point cloud data and data preprocessing

[0034] The present application obtains the original point cloud data of the vehicle's surrounding environment through the radar installed on the top of the autonomous vehicle or evenly installed around the vehicle, and preprocesses the obtained original point cloud data, including the unification of the spatial coordinate system and the time synchronization, to ensure the unification of the point cloud data coordinate system and eliminate the motion distortion between the point cloud data, facilitating subsequent processing.

[0035] The spatial coordinate system unification process is as follows: first, establish the vehicle body coordinate system, taking the position of the center of the vehicle front as the coordinate origin, and the vehicle forward direction as the X-axis positive direction, and the left direction perpendicular to the X-axis as the Y-axis positive direction; according to the installation position of the radar on the vehicle, the coordinate system of each radar installed on the vehicle is converted to the vehicle body coordinate system through rotation and translation, to realize the unification of the point cloud data coordinate system. The time synchronization process is as follows: use the system timestamp or high-frequency positioning information of the inertial navigation system on the vehicle to eliminate the motion distortion between the point cloud data.

[0036] Step 2. Target segmentation, generating target polygon

[0037] ​The pre-processed point cloud data is clustered to achieve target segmentation, and target point cloud data is obtained, and a target polygon is generated based on the point cloud data of each target. Considering that the point cloud data is far away from the vehicle, it will form a feature of being dense near and sparse far away. When the same clustering threshold is used to cluster the point cloud, the point cloud far away from the vehicle may not be accurately clustered because the point cloud is sparse (less point cloud), which is not conducive to accurate target segmentation. Therefore, the point cloud data is segmented by setting a segmentation radius, and each segment of point cloud data is clustered, and different clustering thresholds are set for each segment of point cloud data to make the target segmentation more accurate. Since the closer the target is to the vehicle, the higher the threat level, the clustering threshold is related to the distance of the point cloud from the radar. When the distance from the radar is closer, the clustering threshold is set higher. For example, the target point cloud data is divided into two segments by setting a segmentation radius r. According to the danger level of the target within a certain range to the vehicle, r is set to 3 meters. The point cloud with a radius less than 3 meters from the center point of the radar is clustered using a set first clustering threshold (for example, 0.35 m), and the point cloud with a radius greater than or equal to 3 meters from the center point of the radar is clustered using a set second clustering threshold (for example, 0.55 m). The segmented target point cloud is subjected to convex point searching to generate a target polygon. As other embodiments, the acquired point cloud can be divided into more segments according to the distance of the point cloud from the radar, and the segmentation radius can be set according to the danger level of the target within different distance ranges to the vehicle.

[0038] Step 3. Target polygon area splitting

[0039] Due to the characteristics of the convex hull point, the description of the concave target is not accurate, which can easily cause the convex polygon of the concave object to cover the vehicle body, which does not match the actual situation. Therefore, it is necessary to determine whether the generated target polygon and the vehicle body box have intersection points. If there are intersection points, the number of point clouds in the first area, the second area, and the third area of the target polygon is counted. If the number of point clouds in any of the areas is less than the set simplification edge number, the point clouds in the area are merged into the adjacent area of the area, and a new polygon of the target is formed in different areas. The first, second, and third areas refer to the left, middle, and right areas of the vehicle in the main direction of the vehicle driving direction. The set simplification edge number represents that the number of polygon edges generated after the simplification of the target polygon cannot be less than the set simplification edge number. The simplification edge number can be set according to the actual simplification accuracy requirement. As other embodiments, the generated target polygon can not be subjected to area splitting, and the target polygon generated in step 2 is directly simplified through subsequent processes.

[0040] Step 4. Short edge simplification

[0041] It is judged whether there is a side with a length less than a set length threshold in the target polygon. If there is, the side is regarded as a short side. It is judged whether the intersection point of the extensions of the adjacent sides on both sides of the short side deviates by a degree satisfying a set short side deviation threshold. When the intersection point deviation degree is less than the set short side deviation threshold, it is determined that the short side is deleted. When the intersection point deviation degree is greater than the set short side deviation threshold, the short side is still retained. The deviation degree refers to the proportion of the distance from the intersection point of the extensions of the adjacent sides on both sides of the short side to the short side to the distance from the intersection point to the center point of the polygon. In this embodiment, the set length threshold is 5 cm, that is, the side of the target polygon less than 5 cm is regarded as a short side. As another implementation, the length threshold can be determined according to the size of the target object in the actual environment. In this embodiment, the short side deviation threshold is 25%, and as another implementation, the short side deviation threshold can be determined according to the actual simplification requirement.

[0042] Step 5. Determining reliable sides

[0043] According to the radar coordinates and the coordinates of the vertices of the target polygon, the reliable sides of each target polygon after short side simplification are determined. The specific judgment process of the reliable sides is as follows: as shown in FIG. 5, first, the vertex with the maximum angle and the vertex with the minimum angle in the target polygon are found, and the line segment between the vertex with the maximum angle and the vertex with the minimum angle is regarded as a first vector. The line segments between each vertex of the target polygon and the vertex with the minimum angle are regarded as second vectors. The first vector and the second vector are cross-multiplied. When the cross-multiplication result is greater than 0, the corresponding polygon vertex is a reliable vertex. When the cross-multiplication result is less than 0, the corresponding polygon vertex is an unreliable vertex. The sides between the reliable vertices are regarded as reliable sides. The corresponding diagram of the reliable sides is shown in FIG. 6, for example. Figure 2 Figure 3 The solid line close to the laser radar side is a reliable side.

[0044] If one vertex in a side is a reliable vertex and the other vertex is an unreliable vertex, the line segment between the vertex with the maximum angle and the vertex with the minimum angle (the first vector) is regarded as a boundary line. At this time, whether the side is a reliable side is judged according to the distances of the two vertices from the boundary line. When the distance of the reliable vertex from the boundary line is greater than the distance of the unreliable vertex from the boundary line, the side is regarded as a reliable side, and the remaining sides in the polygon except the reliable sides are regarded as invalid sides. The side composed of the vertex with the maximum angle, the vertex with the minimum angle, and the reliable vertex whose cross-multiplication result with the second vector is less than 0 is a reliable side. Therefore, the side composed of one reliable vertex and one unreliable vertex is not regarded as a reliable side.

[0045] The vertex with the maximum angle refers to the point last scanned by the radar in the scanning process of the target polygon, and the vertex with the minimum angle refers to the point first scanned by the radar in the scanning process of the target polygon.

[0046] Step 6. Reliable side simplification and invalid side simplification​

[0047] The reliable edge simplification process is as follows: as shown in Figure 2 the number of reliable edges in the target polygon and the length of each reliable edge are counted, and a set number of edges with larger reliable edge lengths are found; the reliable edges are extracted according to a set interval number in a set order, and the extracted reliable edges and the set number of longest edges are taken as the initialization simple edges; the initialization simple edges are subjected to a reasonableness verification, and the initialization simple edges and the remaining edges that do not satisfy the reasonableness verification are taken as the simplification result of the reliable edges of the target polygon; wherein the set number is [2, 5], which is related to the number of reliable edges, and the set number increases as the number of reliable edges increases; the set interval number is also related to the number of reliable edges, and the set interval number increases as the number of reliable edges increases. For example, when the number of reliable edges in the target polygon is 8, the set number is 2, when the number of reliable edges is 12, the set number is 3, and the set number gradually increases as the number of reliable edges increases. When the reliable edges are extracted according to the set interval, if the number of reliable edges is 8, the set interval number can be 3, starting from any reliable edge (the longest edge or the shortest edge), the 10 edges are numbered (1-8) in a certain order (clockwise or counterclockwise), and the selected reliable edges are the edges numbered 1, 4, 7, and 10. If the number of reliable edges is 10, the set interval number can be 4, and the numbering (1-10) is still performed in a certain order, and the selected reliable edges are the edges numbered 1, 5, and 9. In this way, as the number of reliable edges increases, the set interval number gradually increases.

[0048] The reasonableness verification process is: the deviation degree of the intersection point of each two adjacent edges in the initialization simple edge is calculated, when the deviation degree of the intersection point of the adjacent edges is greater than the set deviation threshold, the reliable edge between the two adjacent edges is retained; when the deviation degree of the intersection point of the adjacent edges is less than the set deviation threshold, the reliable edge between the two adjacent edges is deleted. Wherein the deviation degree of the intersection point of each two adjacent edges in the initialization simple edge is that the distance from the intersection point to the target point is the proportion of the distance from the intersection point to the center point of the polygon. Since the reliable edge can be one or more, when the reliable edge between the two adjacent edges is one, the target point is the intersection point of the center point of the polygon and the secondary reliable edge, and when the reliable edge between the two adjacent edges is more than one, the target point is the vertex farthest from the center point of the polygon. For example, in this embodiment, the deviation threshold is 25%, and as another embodiment, the deviation threshold can be determined according to the actual simplification requirement.

[0049] The invalid edge simplification process is as follows: Count the number of invalid edges and their lengths, retaining the longest invalid edges (a predetermined number). Since invalid edges are farther from the radar, their accuracy in describing the target is lower than that of valid edges. Selecting the longest invalid edges (a predetermined number) and deleting the others will not significantly affect the overall target polygon. The predetermined number of invalid edges is related to the total number of invalid edges in the target polygon; the more invalid edges there are, the larger the predetermined number should be. For example, when there are 6 invalid edges, select the two longest ones; when there are 10 invalid edges, select the three longest ones.

[0050] The finally extracted reliable edges and invalid edges are combined to form a new target polygon, thus simplifying all target polygons.

[0051] System Implementation Examples

[0052] This invention also proposes a target perception system for autonomous vehicles. The system includes a processor and a memory, the memory storing a computer program executable on the processor. When the processor executes the computer program, it implements the methods described in the above-described method embodiments. That is, the methods in the above-described method embodiments should be understood as a flow of a target perception method for autonomous vehicles that can be implemented by computer program instructions. These computer program instructions can be provided to the processor, causing the processor to execute these instructions to produce the functions specified for implementing the above-described method flow.

[0053] In this embodiment, the processor refers to a processing device such as a microprocessor (MCU) or a programmable logic device (FPGA); the memory refers to a physical device used to store information, which typically involves digitizing the information and then storing it using media that utilizes electrical, magnetic, or optical methods. Examples include: various types of memory that store information using electrical energy, such as RAM and ROM; various types of memory that store information using magnetic energy, such as hard disks, floppy disks, magnetic tapes, magnetic core memory, bubble memory, and USB flash drives; and various types of memory that store information using optical methods, such as CDs or DVDs. Of course, there are other types of memory, such as quantum memories and graphene memories.

[0054] The device comprising the aforementioned memory, processor, and computer program is implemented by the processor executing corresponding program instructions within a computer. The processor can run various operating systems, such as Windows, Linux, Android, and iOS. As another implementation, the system may also include a display for showing the target perception results of the autonomous vehicle for operator reference.

[0055] Vehicle Example

[0056] This invention also proposes an autonomous vehicle, which includes a radar and an inertial navigation system mounted on the vehicle body. The radar can be mounted on the top of the vehicle or evenly distributed around the vehicle to acquire target information within a 360-degree range around the vehicle. The inertial navigation system is located inside the vehicle body to measure the angular velocity and acceleration of objects in three-dimensional space, calculate the object's attitude, and correct motion distortion of the radar point cloud. Both the radar and the inertial navigation system are connected to a controller to implement the target perception method for autonomous vehicles as described in the above method embodiments. This method has been described in detail in the above method embodiments and will not be repeated here.

Claims

1. A method for target perception of an autonomous vehicle, the method comprising: The method comprises the following steps: 1) obtaining original point cloud data of a vehicle surrounding environment by radar, and preprocessing the original point cloud data; the preprocessing comprises unifying space coordinates and time of all original point cloud data; 2) clustering the preprocessed point cloud data to realize target segmentation, obtaining target point cloud data, and generating a target polygon based on the target point cloud data; 3) for each target polygon, determining a reliable edge of each target polygon according to radar coordinates and polygon vertex coordinates; the judgment method of the reliable edge is as follows: finding an angle maximum vertex and an angle minimum vertex in the target polygon, taking a line between the angle maximum vertex and the angle minimum vertex as a first vector, taking lines between each vertex of the target polygon and the angle minimum vertex as second vectors, performing cross multiplication on the first vector and the second vectors, when the cross multiplication result is greater than 0, the corresponding polygon vertex is a reliable vertex, when the cross multiplication result is less than 0, the corresponding polygon vertex is an unreliable vertex, and taking edges between reliable vertices as reliable edges; wherein the angle maximum vertex refers to a point last scanned by the radar in scanning the target polygon, and the angle minimum vertex refers to a point first scanned by the radar in scanning the target polygon; 4) simplifying the reliable edges of each target polygon: counting the number of reliable edges in the target polygon and the length of each reliable edge, finding a set number of edges with the maximum length, extracting the reliable edges according to a set interval number and a set order, taking the extracted reliable edges and the set number of longest edges as initial simplified edges, and performing rationality verification on the initial simplified edges, taking the initial simplified edges and the remaining edges that do not satisfy the rationality verification as the simplification result of the reliable edges of the target polygon; the rationality verification process is as follows: calculating the deviation degree of intersection points of two adjacent edges in the initial simplified edges, when the deviation degree of the intersection points of the two adjacent edges is greater than a set deviation threshold, the reliable edge between the two adjacent edges is retained; when the deviation degree of the intersection points of the two adjacent edges is less than the set deviation threshold, the reliable edge between the two adjacent edges is deleted.

2. The target perception method of an autonomous vehicle according to claim 1, wherein, The method further comprises simplifying invalid edges: counting the number of invalid edges and the length of the edges, and retaining a set number of longest edges in the invalid edges; wherein the invalid edge refers to an edge far from the radar. 3.The target perception method of an autonomous vehicle according to claim 1, wherein, Before determining the reliable edge, the target polygon is simplified by short edges, first taking edges with a length less than a set length threshold in the target polygon as short edges, judging whether the deviation degree of intersection points of extension lines of adjacent edges on both sides of the short edge satisfies a set short edge deviation threshold, when the deviation degree of the intersection points is less than the set short edge deviation threshold, it is determined that the short edge is deleted; when the deviation degree of the intersection points is greater than the set short edge deviation threshold, the short edge is still retained; wherein the deviation degree refers to the proportion of the distance from the intersection point of the extension lines of the adjacent edges on both sides of the short edge to the short edge to the distance from the intersection point to the center point of the polygon. 4.The target perception method of an autonomous vehicle according to any one of claims 1-3, wherein, Before step 3), it is judged whether the target polygon and the vehicle body box exist intersection points, if the intersection points exist, the point cloud number of the target polygon in the first area, the second area and the third area is counted, if the point cloud number in any area is less than the set number of simplified edges, the point cloud in the area is merged to the adjacent area of the area, and a new target polygon is formed in the corresponding area; wherein the first area, the second area and the third area respectively refer to the left part, the middle part and the right part of the vehicle in the direction of vehicle driving. 5.The target perception method of an autonomous vehicle according to any one of claims 1-3, wherein, The point cloud clustering in step 2) adopts segmented clustering, and is divided into at least two segments, and each segment is clustered according to the corresponding set clustering threshold; wherein the set clustering threshold is related to the distance of the point cloud from the radar, and when the point cloud is closer to the radar, the clustering threshold is set smaller. 6.The target perception method of an autonomous vehicle according to claim 1, wherein, The set number in step 4) is [2, 5], which is related to the number of reliable edges, and when the number of reliable edges is more, the set number is larger; the set interval number in step 4) is also related to the number of reliable edges, and when the number of reliable edges is more, the set interval number is larger. 7.The target perception method of an autonomous vehicle according to claim 1, wherein, The deviation degree refers to that the vertex of the reliable edge in the middle of two adjacent edges in the initial simplified edge is taken as a target point, and the distance from the intersection point of the adjacent edge to the target point accounts for the proportion of the distance from the intersection point to the center point of the polygon.

8. A target perception system for an autonomous vehicle, the system comprising: The system comprises a processor and a memory, and the processor is used to process the computer program stored in the memory to realize the target perception method of the automatic driving vehicle according to any one of claims 1-7.

9. An autonomous vehicle, comprising: The vehicle comprises a radar and an inertial navigation system arranged on the vehicle body, and the radar and the inertial navigation system are connected with the controller to realize the target perception method of the automatic driving vehicle according to any one of claims 1-7.

Images