Target identification method and device for laser radar point cloud, and storage medium

Abstract

The embodiment of the invention discloses a laser radar point cloud target identification method and device and a storage medium. The method comprises the following steps: obtaining a distance value matrix corresponding to a first point cloud frame and a second point cloud frame according to the first point cloud frame and the second point cloud frame collected by a laser radar; dividing points of the first point cloud frame and the second point cloud frame according to a first preset threshold to obtain a plurality of point sets of the first point cloud frame and the second point cloud frame; according to the plurality of point sets and the distance value matrix corresponding to the point cloud frame, obtaining the value of the characteristic parameter of each point set; obtaining a first to-be-matched point in the current point cloud frame according to a first preset threshold value and a second preset threshold value; obtaining a current to-be-matched point set according to the first to-be-matched point and the plurality of point sets of the second point cloud frame; judging whether the first to-be-matched point is a noisy point or not according to the first to-be-matched point, the current to-be-matched point set and a plurality of point sets of the first point cloud frame; and performing target identification according to the noisy point judgment result. The method can effectively improve the radar target recognition accuracy.

Description

Technical Field

[0001] The present application relates to the field of laser radar, and specifically to a target recognition method, device and storage medium for laser radar point cloud. Background Art

[0002] LiDAR is an active remote sensing device that uses laser pulses for target detection. Based on the scanning beam emitted by the LiDAR, information such as the target's distance, size, speed, and surface reflectivity can be obtained. Currently, LiDAR is widely used in fields such as autonomous driving, industrial mapping, robotics, and smart transportation. In various application scenarios, ambient light can interfere with the LiDAR's received signal. When the target is absent or the target's reflected signal is weak, noise signals may be output by the detection system to the downstream module, appearing as cluttered noise in the 3D point cloud image, affecting target feature recognition. The sun is the most common source of interfering light. Dense sunlight noise is difficult to distinguish from the point cloud morphology of small, distant targets. During point cloud processing, high levels of filtering make it difficult to preserve detailed information about small, distant targets; low levels of filtering result in a high rate of sunlight noise, which in turn affects LiDAR's detection accuracy. Summary of the Invention

[0003] In order to improve the accuracy of laser radar target recognition, the embodiments of the present application disclose a target recognition method, device and storage medium for laser radar point cloud.

[0004] In a first aspect, an embodiment of the present application provides a method for target recognition of a laser radar point cloud, the method comprising:

[0005] Obtaining a first distance value matrix and a second distance value matrix based on a first point cloud frame and a second point cloud frame acquired by the laser radar, wherein an element of the distance value matrix corresponds one-to-one to a point of the point cloud frame, and the first point cloud frame, the second point cloud frame, and the current point cloud frame are point cloud frames acquired sequentially in time sequence;

[0006] Dividing the points of the first point cloud frame and the points of the second point cloud frame according to a first preset threshold to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame;

[0007] Obtaining a value of a feature parameter of each of the point sets according to the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix, and the second distance value matrix;

[0008] Obtaining at least one to-be-matched point from the current point cloud frame according to the first preset threshold and the second preset threshold, wherein the at least one to-be-matched point includes a first to-be-matched point;

[0009] Obtaining, according to the first to-be-matched point, position coordinates of elements of a distance value matrix corresponding to the first to-be-matched point;

[0010] Obtaining at least one to-be-matched point set of the second point cloud frame according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched points and the multiple point sets of the second point cloud frame;

[0011] Calculating a first absolute value, wherein the first absolute value is the absolute value of the difference between the value of the feature parameter of the current set of points to be matched in the second point cloud frame and the value of the element corresponding to the first set of points to be matched;

[0012] When the first absolute value is less than a third preset threshold, determining whether the first to-be-matched point is a noise point according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched point, the values of the characteristic parameters of the current to-be-matched point set, and the values of the characteristic parameters of the multiple point sets of the first point cloud frame;

[0013] Target recognition is performed based on the noise point judgment result of the current point cloud frame and the first point to be matched.

[0014] In some embodiments, the points of the first point cloud frame and the points of the second point cloud frame are divided according to a first preset threshold to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame, including: according to the distance value matrix corresponding to each of the point cloud frames, obtaining the value of the adjacent first element and the value of the second element in the distance value matrix corresponding to each of the point cloud frames; when the absolute value of the difference between the value of the first element and the value of the second element is less than the first preset threshold, the points corresponding to the first element and the points corresponding to the second element are divided into the same point set; or when the absolute value of the difference between the value of the first element and the value of the second element is greater than or equal to the first preset threshold, the points corresponding to the first element and the points corresponding to the second element are divided into different point sets.

[0015] In some embodiments, the feature parameters are a combination of one or more of the following parameters: average value, centroid position, fitted normal vector coefficient, flatness ratio or roundness; accordingly, the value of the feature parameter of each point set is obtained based on the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix and the second distance value matrix, including: obtaining the average value of each point set based on the value of the element corresponding to the points included in each point set; or obtaining the value of the centroid position of each point set based on the average value of the position coordinates of the element corresponding to the points included in each point set; or obtaining the value of the fitted normal vector coefficient of each point set based on the value of the element corresponding to the points included in each point set and the position coordinates of the element corresponding to the points included in each point set; or obtaining the value of the flatness ratio of each point set based on the extreme difference of the position coordinates of the element corresponding to the points included in each point set; or obtaining the value of the roundness of each point set based on the position coordinates of the element corresponding to the points included in each point set and the centroid position of each point set.

[0016] In some embodiments, obtaining at least one point to be matched from the current point cloud frame based on the first preset threshold and the second preset threshold includes: obtaining a first number of elements based on the elements of the distance value matrix corresponding to a point of the current point cloud frame and a first preset neighborhood, wherein the first number of elements is the number of elements within the first preset neighborhood whose absolute value of the difference between the value of the element corresponding to a point of the current point cloud frame is less than the first preset threshold; when the first number of elements is greater than the second preset threshold, determining a point of the current point cloud frame as the point to be matched.

[0017] In some embodiments, the method of obtaining at least one point set to be matched in the second point cloud frame based on the position coordinates of the elements of the distance value matrix corresponding to the first point to be matched and the multiple point sets of the second point cloud frame includes: obtaining the position coordinates of the element corresponding to a point of the second point cloud frame based on the position coordinates of the element of the distance value matrix corresponding to the first point to be matched; obtaining the position coordinates of multiple elements within the second preset neighborhood based on the position coordinates of the element of the distance value matrix corresponding to a point of the second point cloud frame and the second preset neighborhood; obtaining at least one point set to be matched based on the position coordinates of multiple elements within the second preset neighborhood and the multiple point sets of the second point cloud frame, wherein the point set to be matched is the point set including at least one point corresponding to the element within the second preset neighborhood.

[0018] In some embodiments, determining whether the first to-be-matched point is a noise point according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched point, the values of the characteristic parameters of the current to-be-matched point set, and the values of the characteristic parameters of the multiple point sets of the first point cloud frame includes:

[0019] Obtaining an inter-frame distance difference according to an absolute value of a difference between the average value of each point set of the first point cloud frame and the average value of the current point set to be matched;

[0020] Obtaining an inter-frame tilt angle difference according to the value of the centroid position of each point set of the first point cloud frame, the value of the centroid position of the current point set to be matched, and the position coordinates of the element of the distance value matrix corresponding to the first point to be matched;

[0021] Obtaining an inter-frame normal vector deviation according to the value of the fitted normal vector coefficient of each point set of the first point cloud frame and the value of the fitted normal vector coefficient of the current point set to be matched;

[0022] Obtaining an inter-frame divergence difference according to the number of points and the average value of each point set of the first point cloud frame and the number of points and the average value of the current point set to be matched;

[0023] Obtaining an inter-frame flatness ratio difference according to the flatness ratio value of each point set of the first point cloud frame and the flatness ratio value of the current point set to be matched;

[0024] Obtaining an inter-frame roundness difference according to the roundness value of each point set of the first point cloud frame and the roundness value of the current point set to be matched;

[0025] When there is no point set among the multiple point sets of the first point cloud frame, so that the inter-frame distance difference is less than the third preset threshold, the inter-frame inclination difference is less than the fourth preset threshold, the inter-frame normal vector deviation is less than the fifth preset threshold, the inter-frame divergence difference is less than the sixth preset threshold, the inter-frame flatness ratio difference is less than the seventh preset threshold, and the inter-frame roundness difference is less than the eighth preset threshold, the first point to be matched is determined to be the noise point.

[0026] In some embodiments, the method further includes: calculating the first absolute value corresponding to each of the set of points to be matched in the second point cloud frame; when the first absolute value corresponding to each of the set of points to be matched is greater than or equal to the third preset threshold, the first point to be matched is a noise point.

[0027] In a second aspect, an embodiment of the present application provides a target recognition device for a laser radar point cloud, comprising:

[0028] an acquisition module, configured to obtain a first distance value matrix and a second distance value matrix based on a first point cloud frame and a second point cloud frame acquired by the laser radar, wherein an element of each distance value matrix corresponds one-to-one to a point of each point cloud frame, and the first point cloud frame, the second point cloud frame, and the current point cloud frame are point cloud frames acquired sequentially in time sequence;

[0029] a point cloud processing module, configured to divide the points of the first point cloud frame and the points of the second point cloud frame according to a first preset threshold value, to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame; the point cloud processing module is further configured to obtain a value of a characteristic parameter of each point set based on the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix, and the second distance value matrix; the point cloud processing module is further configured to obtain at least one point to be matched from the current point cloud frame based on the first preset threshold value and the second preset threshold value, wherein the at least one point to be matched includes a first point to be matched; the point cloud processing module is further configured to obtain, based on the first point to be matched, the position coordinates of the elements of the distance value matrix corresponding to the first point to be matched; the point cloud processing module is further configured to obtain, based on the position coordinates of the elements of the distance value matrix corresponding to the first point to be matched and the multiple point sets of the second point cloud frame, at least one point set to be matched in the second point cloud frame;

[0030] a noise point identification module, the noise point identification module being configured to calculate a first absolute value, wherein the first absolute value is the absolute value of the difference between a value of a characteristic parameter of a current set of points to be matched in the second point cloud frame and a value of the element corresponding to the first point to be matched, the noise point identification module being further configured to determine whether the first point to be matched is a noise point based on the position coordinates of an element of a distance value matrix corresponding to the first point to be matched, the value of the characteristic parameter of the current set of points to be matched, and the values of the characteristic parameters of the multiple point sets in the first point cloud frame when the first absolute value is less than a third preset threshold;

[0031] A target recognition module is used to perform target recognition based on the current point cloud frame and the noise point judgment result of the first point to be matched.

[0032] The implementation of the device embodiment of the present invention can refer to the above method embodiment.

[0033] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to implement the steps of the target recognition method of the lidar point cloud in the above embodiment.

[0034] This application discloses a method, device, and storage medium for target recognition using a LiDAR point cloud. This method performs multiple frame-to-frame matching on the first, second, and current point cloud frames continuously acquired by the LiDAR, thereby filtering out sunlight noise from the current point cloud frame. This method can preserve point cloud information for small, distant targets, improving the accuracy of LiDAR recognition of distant targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application.

[0036] Figure 1 This is a flowchart of a target recognition method for a laser radar point cloud provided in an embodiment of the present application;

[0037] Figure 2 This is a flowchart of a target recognition method for a laser radar point cloud provided in an embodiment of the present application;

[0038] Figure 3 This is a schematic diagram of the movement path of a small target at a distance between three consecutive point cloud frames provided in an embodiment of the present application;

[0039] Figure 4 This is a schematic diagram of the functional modules of a target recognition device for a laser radar point cloud provided in an embodiment of the present application.

[0040] Explanation of reference numerals: 210, acquisition module; 220, point cloud processing module; 230, noise recognition module; 240, target recognition module. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of this application more apparent, embodiments of the present application will be further described in detail below with reference to the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, identical numbers in different drawings represent identical or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Instead, they are merely examples of structures consistent with certain aspects of this application, as detailed in the appended claims.

[0042] LiDAR is susceptible to interference from external light sources during point cloud acquisition, impacting point cloud quality and ranging accuracy. Sunlight is the most common interfering light source. During long-range target detection, noise in the point cloud is primarily sunlight noise caused by interference from sunlight. Some dense sunlight noise can be difficult to distinguish from the point cloud morphology of small targets based on local information. During point cloud processing, excessive filtering can result in the removal of small, distant targets or the loss of some texture information. Low filtering results in a high rate of residual sunlight noise, which can interfere with target detection and reduce LiDAR detection accuracy.

[0043] This application discloses a target recognition method for laser radar point cloud. By multi-frame matching, sunlight noise is distinguished from the target point cloud, thereby filtering out sunlight noise and retaining the point cloud information of small targets at a distance. The flow chart of this method is as follows Figure 1 The specific steps are as follows:

[0044] In some embodiments, steps S10-S30 are first executed to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame as well as the values of the characteristic parameters of each point set, which are used for the noise discrimination process of the current point cloud frame in steps S40-S60.

[0045] S10. Obtain a first distance value matrix and a second distance value matrix according to the first point cloud frame and the second point cloud frame collected by the laser radar.

[0046] In one embodiment, based on the point cloud data of the laser radar, the distance data of multiple point cloud frames collected continuously are obtained. The multiple point cloud frames include a first point cloud frame, a second point cloud frame and a current point cloud frame collected in sequence. According to the distance data of the first point cloud frame, a first distance value matrix is obtained. According to the distance data of the second point cloud frame, a second distance value matrix is obtained. According to the distance data of the current point cloud frame, a distance value matrix corresponding to the current point cloud frame is obtained. The elements of a distance value matrix correspond one-to-one to the points of a point cloud frame. In one example, the distance value matrices corresponding to each point cloud frame have the same matrix size and include M×N elements for recording the distance data of each point cloud frame. Wherein, M is the number of rows of the distance value matrix, and N is the number of columns of the distance value matrix. In some embodiments, the laser radar includes a mechanical laser radar, a semi-solid laser radar, a hybrid scanning laser radar and a pure solid-state laser radar.

[0047] S20. Divide the points of the first point cloud frame and the points of the second point cloud frame according to a first preset threshold to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame.

[0048] In one embodiment, the values of adjacent first elements and second elements in each distance value matrix are obtained based on the first distance value matrix and the second distance value matrix. When the absolute value of the difference between the value of the first element and the value of the second element is less than a first preset threshold, the points corresponding to the first element and the points corresponding to the second element are divided into the same point set; or when the absolute value of the difference between the value of the first element and the value of the second element is greater than or equal to the first preset threshold, the points corresponding to the first element and the points corresponding to the second element are divided into different point sets.

[0049] Taking the processing of the first point cloud frame as an example, in one embodiment, the elements of each column of the first distance value matrix are traversed by column or the elements of each row of the first distance value matrix are traversed by row to divide each point of the first point cloud frame to obtain multiple point sets of the first point cloud frame. In one example, the zero-initialized first connection graph matrix is used to record the point set division results of the point cloud frame corresponding to the first distance value matrix. The elements of a first connection graph matrix correspond one-to-one to the elements of a first distance value matrix. (i, j) is used to represent the position coordinates of the elements in the first distance value matrix, and the value of the element (i, j) is not 0, where i is the matrix row number of the element, j is the matrix column number of the element, i≤M, j≤N. According to the element (1,1) and the first preset neighborhood centered on the element (1,1), the values of multiple elements (m,n) of the first preset neighborhood are obtained, m≤M, n≤N, where the size of the first preset neighborhood can be set to 3×3 or 4×4. When the absolute value of the difference between the value of element (m,n) and the value of element (1,1) is less than a first preset threshold, the points corresponding to element (m,n) and the points corresponding to element (1,1) are divided into the same point set, the sequence number of this point set is marked as 1, and the sequence number value is stored as the connection value in the (1,1) and (m,n) positions of the first connection graph matrix. When the absolute value of the difference between the value of element (m,n) and the value of element (1,1) is greater than or equal to the first preset threshold, the points corresponding to element (1,1) are divided into a point set, the sequence number of this point set is marked as 1, and the sequence number value is stored as the connection value in the (1,1) position of the first connection graph matrix. The point corresponding to element (m,n) belongs to another point set. The other point set to which the point corresponding to element (m,n) belongs is sequenced in the traversal order, and the sequence number value is stored as the connection value in the (m,n) position of the first connection graph matrix. Repeat the above point set partitioning process for element (2, 1), element (3, 1), element (4, 1), and finally element (M, 1) to complete the column traversal of the elements in the first column of the first distance value matrix. After the column traversal of the elements in the first column is completed, the column traversal is performed column by column until the N columns are completed, thereby completing the point set partitioning for each point in the first point cloud frame. Repeat the above point set partitioning process for the second point cloud frame to obtain multiple point sets in the second point cloud frame.

[0050] During the above sliding traversal process, when the connection value corresponding to element (m,n) is already stored in the first connection graph matrix, and the absolute value of the difference between the value of element (i,j) and the value of element (m,n) is less than a first preset threshold, the connection value corresponding to element (m,n) is used as the connection value corresponding to element (i,j) and stored at position (i,j) in the first connection graph matrix. When the absolute value of the difference between the value of element (i,j) and the value of element (m,n) is greater than or equal to the first preset threshold, the point corresponding to element (i,j) belongs to a new point set, and the new point set is serially numbered according to the traversal order to obtain the connection value corresponding to element (i,j), and the connection value is stored at position (i,j) in the first connection graph matrix.

[0051] In one embodiment, the first preset threshold is set to a fixed value, such as 0.08m, 0.1m, or 0.12m. In another embodiment, the first preset threshold can be set in segments based on the distance between the target and the lidar. When the distance range is 100m-129m, the first preset threshold is set to 0.08m; when the distance range is 130m-179m, the first preset threshold is set to 0.1m; and when the distance range is 180m-250m, the first preset threshold is set to 0.12m.

[0052] Before executing step S30 for the first point cloud frame and the second point cloud frame, the characteristic parameter matrix is initialized to obtain the initialized characteristic parameter matrix, and the initialization of the characteristic parameter matrix includes: initializing a one-dimensional matrix MaxRow for recording the maximum matrix row number of the elements corresponding to the points in each point set, a one-dimensional matrix MaxCol for recording the maximum matrix column number of the elements corresponding to the points in each point set, a one-dimensional matrix CenRow for recording the centroid row number of each point set, a one-dimensional matrix CenCol for recording the centroid column number of each point set, and a one-dimensional matrix CenRow for recording the centroid column number of each point set. a one-dimensional matrix ObjDist for recording the average values of the elements corresponding to the points, a one-dimensional matrix AreaCnt for recording the number of points included in each point set, a one-dimensional matrix RowCoef for recording the row coefficients of the fitted normal vector of each point set, a one-dimensional matrix ColCoef for recording the column coefficients of the fitted normal vector of each point set, a one-dimensional matrix FlatRio for recording the flatness ratio of each point set, a one-dimensional matrix CirDeg for recording the roundness of each point set, and a one-dimensional matrix CalDone for recording whether the statistics of the characteristic parameters of each point set are completed.

[0053] S30. Obtain a value of a characteristic parameter of each point set according to the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix, and the second distance value matrix.

[0054] Taking the first point cloud frame as an example, in one embodiment, the multiple point sets of the first point cloud frame include a point set k with a serial number of k. The characteristic parameters of point set k are obtained based on the points included in point set k, the values of the elements corresponding to the points included in point set k, and the position coordinates of the elements corresponding to the points included in point set k. In one example, the number of points included in point set k is recorded in the kth row of the one-dimensional matrix AreaCnt. The maximum matrix row number of the elements corresponding to the points included in point set k is used as the maximum row number of point set k and recorded in the kth row of the one-dimensional matrix MaxRow. The maximum matrix column number of the elements corresponding to the points included in point set k is used as the maximum column number of point set k and recorded in the kth row of the one-dimensional matrix MaxCol. The average value of the values of the elements corresponding to the points included in point set k is recorded in the kth row of the one-dimensional matrix ObjDist. The average value of the matrix row numbers of the elements corresponding to the points included in point set k is used as the matrix row number of the centroid and recorded in the kth row of the one-dimensional matrix CenRow. The average of the matrix column numbers of the elements corresponding to the points in point set k is recorded as the matrix column number of the centroid, and is recorded in the kth row of the one-dimensional matrix CenCol. The matrix row and column numbers of the centroid are used as the centroid position (CenRow(k), CenCol(k)). The kth position of the feature statistics completion flag CalDone, the fitted normal vector row coefficient RowCoef, the fitted normal vector column coefficient ColCoef, the flatness ratio FlatRio, the roundness CirDeg, etc. is written to 0.

[0055] When the value AreaCnt(k) in the k-th row of the one-dimensional matrix AreaCnt is less than the point count threshold ClassNumTh, reset all positions in the first connectivity graph matrix with a connectivity value of k to 0. Decrease the connectivity value of all positions in the first connectivity graph matrix with a connectivity value greater than k by 1. Delete the k-th position of the one-dimensional matrices MaxRow, CenRow, MaxCol, CenCol, AreaCnt, CalDone, ObjDist, RowCoef, ColCoef, FlatRio, and CirDeg.

[0056] When AreaCnt(k) is greater than or equal to the point number threshold ClassNumTh, the range of the row numbers of the elements corresponding to the points included in the point set k and the range of the column numbers of the elements corresponding to the points included in the point set k are calculated, and the ratio of the range of the column numbers to the range of the row numbers is used as the flatness ratio of the point set and recorded in the kth row of the one-dimensional matrix FlatRio. When the range of the row number is 0, the range of the column number is recorded. The value of the element corresponding to each point in the point set k is used as the dependent variable, and the row number of the element corresponding to each point and the column number of the element corresponding to each point are used as independent variables. A binary least squares regression calculation is performed, and the obtained row variable coefficient is used as the row coefficient of the fitted normal vector RowCoef(k) of the point set k, and the column variable coefficient is used as the column coefficient of the fitted normal vector ColCoef(k) of the point set k. Calculate the fitting distance d from the element (i, j) corresponding to the point in the point set k to the centroid position (CerRow(k), CerCol(k)) of the point set k. k :

[0057]

[0058] d k The ratio of the maximum value to the minimum value is taken as the roundness CirDeg(k) of point set k, and the value of the feature statistics completion mark CalDone(k) is recorded as 1 to complete the feature parameter statistics of point set k.

[0059] After the feature parameter statistics for point set k are complete, the elements in the one-dimensional matrix CalDone are traversed. When the values of all elements in the one-dimensional matrix CalDone are 1, it is determined that the feature parameter statistics for multiple point sets in the point cloud frame have been completed. Otherwise, the position coordinates of the elements with a value of 0 in the one-dimensional matrix CalDone are used to obtain the sequence number of each point set for which feature parameter statistics have not been completed, and feature parameter statistics are repeated. The above operation is repeated for the second point cloud frame to obtain the value of the feature parameter of each point set in the second point cloud frame.

[0060] In some embodiments, based on the characteristic parameter statistics of the first point cloud frame and the characteristic parameter statistics of the second point cloud frame, steps S40-S70 are executed for the current point cloud frame to perform inter-frame matching and target recognition to distinguish between noise points and target point clouds in the current point cloud frame.

[0061] S40. Obtain at least one point to be matched from the current point cloud frame according to the first preset threshold and the second preset threshold.

[0062] In one embodiment, a first number of elements is obtained based on an element corresponding to a point in the current point cloud frame and a first preset neighborhood, where the first number of elements is the number of elements within the first preset neighborhood whose absolute value of the difference between the value of the element corresponding to the point in the current point cloud frame is less than a first preset threshold. When the first number of elements is greater than a second preset threshold, a point in the current point cloud frame is determined to be a point to be matched. When the first number of elements is less than or equal to the second preset threshold, other points in the current point cloud frame are traversed to determine whether other points in the current point cloud frame are points to be matched.

[0063] In one example, based on the element (i, j) corresponding to a point in the current point cloud frame and a first preset neighborhood centered on element (i, j), the values of multiple elements (m, n) within the first preset neighborhood are obtained, where the size of the first preset neighborhood can be set to 3×3 or 4×4. When the absolute value of the difference between the value of element (m, n) and the value of element (i, j) is less than a first preset threshold, element (m, n) is recorded as a valid element. The multiple elements (m, n) within the first preset neighborhood are traversed to obtain the number of valid elements. When the number of valid elements is greater than a second preset threshold, the point corresponding to element (i, j) is a to-be-matched point in the current point cloud frame, and step S50 is performed for this to-be-matched point. When the number of valid elements is greater than the second preset threshold, it is determined that the point corresponding to element (i, j) is not a to-be-matched point. Based on the first and second preset thresholds, it is determined whether the point corresponding to element (i+1, j) or the point corresponding to element (i, j+1) is a to-be-matched point, where i+1≤M and j+1≤N.

[0064] S50. Obtain at least one point set to be matched in the second point cloud frame according to the first point set to be matched and the multiple point sets in the second point cloud frame.

[0065] In one embodiment, the at least one point to be matched obtained in step S40 includes a first point to be matched, and the position coordinates of the element corresponding to the first point to be matched are obtained based on the first point to be matched. Based on the position coordinates of the element corresponding to the first point to be matched, the position coordinates of the element corresponding to a point in the second point cloud frame are obtained, wherein the position coordinates of the element corresponding to a point in the second point cloud frame are equal to the position coordinates of the element corresponding to the first point to be matched. Based on the position coordinates of the element corresponding to a point in the second point cloud frame and a second preset neighborhood (size of 5×5 or 6×6), the position coordinates of multiple elements in the second preset neighborhood are obtained. Based on the position coordinates of multiple elements in the second preset neighborhood and multiple point sets of the second point cloud frame, at least one point set to be matched is obtained, wherein the point set to be matched is a point set in the multiple point sets of the second point cloud frame that includes points corresponding to elements of at least one second preset neighborhood.

[0066] S60: Determine whether the first to-be-matched point is a noise point based on the first to-be-matched point, the to-be-matched point set of the second point cloud frame, and the multiple point sets of the first point cloud frame.

[0067] In one embodiment, the average value of each to-be-matched point set is obtained based on the sequence number of the to-be-matched point set in the second point cloud frame and the one-dimensional matrix ObjDist2. The first absolute value corresponding to each to-be-matched point set in the second point cloud frame is calculated. When the first absolute value corresponding to each to-be-matched point set is greater than or equal to a third preset threshold, it is determined that each to-be-matched point set does not meet the first inter-frame matching condition, and the first to-be-matched point is determined to be a noise point. When the first absolute value is less than the third preset threshold, it is determined that the current to-be-matched point set meets the first inter-frame matching condition. Then, all point sets in the first point cloud frame are traversed to determine whether there is a point set in the first point cloud frame that meets the second inter-frame matching condition. The second inter-frame matching condition is set based on the position coordinates of the element corresponding to the first to-be-matched point, the value of the feature parameter of the current to-be-matched point set that meets the first inter-frame matching condition, and the value of the feature parameter of multiple point sets in the first point cloud frame. When each point set in the first point cloud frame does not meet the second inter-frame matching condition, the first to-be-matched point is determined to be a noise point. When a point set in the first point cloud frame satisfies the second inter-frame matching condition, it is determined that the first point to be matched is not a noise point, and the process returns to step S40 to process other points in the current point cloud frame.

[0068] In one embodiment, the third preset threshold can be set according to the distance between the target and the lidar, the radar frame rate, etc. When the distance range is 130m-179m and the radar frame rate is 20Hz, the third preset threshold is set to 0.12m; when the distance range is 180m-250m and the radar frame rate is 20Hz, the third preset threshold is set to 0.15m; when the distance range is 180m-250m and the radar frame rate is 10Hz, the third preset threshold is set to 0.20m.

[0069] In one example, for a point set numbered S1 in the first point cloud frame, the current set of points to be matched that meet the first inter-frame matching condition is recorded as matching point set S2, and the element coordinates corresponding to the first point to be matched are marked as (i, j). The second inter-frame matching condition includes the following six sub-conditions:

[0070] Subcondition 1: The inter-frame distance difference is less than a third preset threshold, wherein the inter-frame distance difference is the absolute value of the difference between the average value ObjDist1(S1) of the point set S1 of the first point cloud frame and the average value ObjDist2(S2) of the matching point set S2 of the second point cloud frame.

[0071] Sub-condition 2: The inter-frame tilt difference is less than a fourth preset threshold value DipDiffTh, where the inter-frame tilt difference satisfies the following relationship, where the denominator is 1 when it is 0:

[0072]

[0073] Among them, CenRow1(S1) is the centroid row number of point set S1, CenCol1(S1) is the centroid column number of point set S1, CenRow1(S2) is the centroid row number of matching point set S2, and CenCol2(S2) is the centroid column number of matching point set S2.

[0074] Sub-condition 3: The inter-frame normal vector deviation is less than a fifth preset threshold DirDiffTh, where the inter-frame normal vector deviation satisfies the following relationship:

[0075] |RowCoef1(S1)×ColCoef2(S2)-ColCoef1(S1)×RowCoef2(S2)| <DirDiffTh

[0076] Among them, RowCoef1(S1) is the row coefficient of the fitted normal vector of point set S1, ColCoef1(S1) is the column coefficient of the fitted normal vector of point set S1, RowCoef2(S2) is the row coefficient of the fitted normal vector of the matching point set S2, and ColCoef2(S2) is the column coefficient of the fitted normal vector of the matching point set S2.

[0077] Sub-condition 4: The inter-frame divergence difference is less than a sixth preset threshold DivDiffTh, where the inter-frame divergence difference satisfies the following relationship:

[0078]

[0079] Among them, AreaCnt1(S1) is the number of points in point set S1, ObjDist1(S1) is the average value of point set S1, AreaCnt2(S2) is the number of points in matching point set S2, and ObjDist2(S2) is the average value of matching point set S2.

[0080] Sub-condition 5: the absolute value of the difference between the flatness ratio FlatRio1 ( S1 ) of the point set S1 and the flatness ratio FlatRio2 ( S2 ) of the matching point set S2 is less than a seventh preset threshold.

[0081] Sub-condition 6: the absolute value of the difference between the circularity CirDeg1 ( S1 ) of the point set S1 and the circularity CirDeg2 ( S2 ) of the matching point set S2 is less than an eighth preset threshold.

[0082] Traverse the point sets of the first point cloud frame. If no point set S1 in the first point cloud frame satisfies sub-conditions 1 to 6 at the same time, determine that the first point to be matched is a noise point. Otherwise, return to step S40 to process other points in the current point cloud frame.

[0083] During the detection of small targets at long distances by LiDAR, the noise in the point cloud is mainly sunlight noise caused by the interference of sunlight. The point cloud of sunlight noise is scattered, amorphous, and has low inter-frame correlation. Figure 3 The points in the point cloud corresponding to the target are relatively concentrated, have a fixed shape, and exhibit a smooth variation across multiple frames. d1, d2, and d0 correspond to the distance values of the target in the first, second, and current point cloud frames, respectively. Based on the distinction between noise and target point clouds, the target point clouds in consecutively acquired point cloud frames satisfy the following requirements: the positional variation of the target's center of mass is limited; the direction vector of the target's center of mass relative to the radar is nearly parallel; the number of points in the target is proportional to the square of the distance from the target's center of mass to the radar; the angle between the normal vectors fitted to the hyperplane of the target point cloud is limited; and the shape of the target point cloud remains essentially unchanged. Therefore, in the second inter-frame matching criteria, sub-condition 1 constrains the distance variation of the target point cloud's center of mass; sub-condition 2 constrains the positional variation of the target point cloud's center of mass; sub-condition 3 constrains the direction of the fitted normal vector of the target point cloud; and sub-conditions 4 through 6 constrain the divergence, flatness ratio, and roundness of the point cloud shape. Based on these multiple inter-frame matching dimensions, sunlight noise can be effectively distinguished from points included in the target point cloud.

[0084] S70: Perform target recognition based on the current point cloud frame and the noise determination result of the first point to be matched.

[0085] After noise points are identified for all points in the current point cloud frame, the noise points are deleted from the current point cloud frame to update the current point cloud frame, resulting in an updated current point cloud frame. Target recognition is performed based on the updated current point cloud frame to obtain target point cloud information. In some embodiments, the target point cloud information includes one or more combinations of parameters such as the number of points in the target point cloud, the spatial coordinates of the target's center of mass, the distance of the target from the lidar, the target's shape and outline, or the target classification.

[0086] In one embodiment, steps S40-S60 can be performed simultaneously during the feature parameter statistics of the current point cloud frame, and the feature parameter statistics of the current point cloud frame are used in the noise point identification process of the point cloud frame after the current point cloud frame, thereby improving the point cloud processing efficiency. In one example, the elements in the distance value matrix corresponding to the current point cloud frame are traversed in a sliding manner to complete the feature parameter statistics of the current point cloud frame. With an element (i, j) of the current point cloud frame whose value is not 0 as the center, a first preset neighborhood of size H×L (3×3) is selected. First, based on steps S40-S60, it is determined whether the point corresponding to the element (i, j) is a noise point. When the point corresponding to the element (i, j) is a noise point, feature parameter statistics are performed on the element (i+1, j) or the element (i, j+1), where i+1≤M, j+1≤N. When the point corresponding to element (i, j) is not a noise point, traverse the elements (m, n) within the first preset neighborhood. When the absolute value of the difference between the values of element (m, n) and element (i, j) is less than the first preset threshold, the points corresponding to element (m, n) and element (i, j) belong to the same point set. The connection value p corresponding to element (m, n) is recorded as the connection value corresponding to element (i, j) at position (i, j) in the connection graph matrix corresponding to the current point cloud frame. The row and column numbers of element (i, j) are stored in the p-th position of the one-dimensional matrices MaxRow and MaxCol, and the value of the p-th position of the one-dimensional matrix AreaCnt is incremented by 1. When the absolute value of the difference between the values of element (m, n) and element (i, j) within the first preset neighborhood is greater than or equal to the first preset threshold, the point corresponding to element (i, j) belongs to the new point set. The total number of point sets q is incremented by 1. The connection value of position (i, j) in the connection graph matrix corresponding to the current point cloud frame is recorded as q+1. The q+1th position of the one-dimensional matrices MaxRow, CenRow, MaxCol, and CenCol corresponds to the row or column number (i, j). The q+1th position of the one-dimensional matrix AreaCnt is set to 1. The q+1th position of the one-dimensional matrices CalDone, ObjDist, RowCoef, ColCoef, FlatRio, and CirDeg is set to 0. Feature parameter statistics are performed on the element (i+1, j) or the element (i, j+1) until the point set partitioning of all points in the current point cloud frame is completed, where i+1≤M and j+1≤N.

[0087] Traverse all point sets in the current point cloud frame. For a point set r in the current point cloud frame, when CalDown(r) is 0, and the row number i minus MaxRow(r) of the traversed current point cloud frame is greater than the statistical delay row threshold H / 2, and the column number j minus MaxCol(r) of the traversed current point cloud frame is greater than the statistical delay column threshold L / 2; or when the row number i of the current point cloud frame is equal to H and the column number j of the current point cloud frame is equal to L, distinguish point set r based on the point number threshold ClassNumTh. When AreaCnt(r) is less than the point number threshold ClassNumTh, reset all positions with a connection value of r in the connection graph matrix corresponding to the current point cloud frame to 0, and reduce the connection value of all positions with a connection value greater than r by 1. Delete the rth position of the one-dimensional matrices such as MaxRow, CenRow, MaxCol, CenCol, AreaCnt, CalDone, ObjDist, RowCoef, ColCoef, FlatRio, and CirDeg. When AreaCnt(r) is greater than or equal to the point count threshold ClassNumTh, the values of the characteristic parameters of point set r, such as the flatness ratio FlatRio(r), the fitted normal vector row coefficient RowCoef(r), the fitted normal vector column coefficient ColCoef(r), and the roundness CirDeg(r), are calculated based on step S30. The value of CalDone(r) is recorded as 1. After traversing the elements of the distance value matrix corresponding to the current point cloud frame, all noise points in the current point cloud frame are deleted to update the current point cloud frame, obtaining the updated current point cloud frame. Target recognition is performed based on the updated current point cloud frame to obtain the target's point cloud information.

[0088] like Figure 4 As shown, an embodiment of the present application provides a target recognition device for a laser radar point cloud, the device comprising:

[0089] An acquisition module 210 is configured to obtain a plurality of distance value matrices based on a plurality of point cloud frames acquired by the laser radar, wherein an element of each distance value matrix corresponds one-to-one to a point of each point cloud frame, and the plurality of point cloud frames include a first point cloud frame, a second point cloud frame, and a current point cloud frame acquired sequentially;

[0090] The point cloud processing module 220 is configured to divide the points of each point cloud frame according to a first preset threshold value to obtain multiple point sets for each point cloud frame. The point cloud processing module is further configured to obtain characteristic parameters of each point set based on the multiple point sets and the multiple distance value matrices. The point cloud processing module is further configured to obtain at least one to-be-matched point in the current point cloud frame based on the first preset threshold value and the second preset threshold value, wherein the at least one to-be-matched point includes the first to-be-matched point. The point cloud processing module is further configured to obtain at least one to-be-matched point set in the second point cloud frame based on the first to-be-matched point in the current point cloud frame and the multiple point sets of the second point cloud frame.

[0091] A noise recognition module 230, the noise recognition module is used to determine that a set of points to be matched in the second point cloud frame meets the first inter-frame matching condition when the absolute value of the difference between the value of a characteristic parameter of a set of points to be matched in the second point cloud frame and the value of the element corresponding to the first point to be matched is less than a third preset threshold value. The noise recognition module is also used to determine that the first point to be matched is a noise point when each set of points to be matched in the second point cloud frame does not meet the first inter-frame matching condition. The noise recognition module is also used to determine that the first point to be matched is a noise point when multiple point sets in the first point cloud frame do not meet the second inter-frame matching condition, wherein the second inter-frame matching condition is set according to the position coordinates of the element corresponding to the first point to be matched, the characteristic parameters of a set of points to be matched that meet the first inter-frame matching condition, and the characteristic parameters of multiple point sets in the first point cloud frame;

[0092] The target recognition module 240 is used to perform target recognition based on the current point cloud frame and noise points.

[0093] In one embodiment, the present application provides a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of steps S10 to S70 in the embodiments of the present application.

[0094] Throughout the description of this application, it should be understood that, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this application belongs. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "and / or" and "and / or" as used herein describe an association relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship. The singular forms "a" and "an" are intended to include the plural forms unless the context clearly indicates otherwise. When used in this specification, the terms "comprising" and / or "including" specify the presence of the recited features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof, i.e., include any and all combinations of one or more of the related listed items. Ordinal numbers such as "first" and "second" cited in the embodiments of this application are merely identifiers and do not denote a particular order or imply relative importance.

[0095] In the present application, unless otherwise expressly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Moreover, a first feature being "above," "above," and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances. The "one or more embodiments" used herein do not refer to the same embodiment, but rather to a combination of specific features, structures, or characteristics in any appropriate manner. The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A target recognition method for laser radar point cloud, characterized in that: include: Obtaining a first distance value matrix and a second distance value matrix based on a first point cloud frame and a second point cloud frame acquired by the laser radar, wherein an element of the distance value matrix corresponds one-to-one to a point of the point cloud frame, and the first point cloud frame, the second point cloud frame, and the current point cloud frame are point cloud frames acquired sequentially in time sequence; Dividing the points of the first point cloud frame and the points of the second point cloud frame according to a first preset threshold to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame; Obtaining a value of a feature parameter of each of the point sets according to the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix, and the second distance value matrix; Obtaining at least one to-be-matched point from the current point cloud frame according to the first preset threshold and the second preset threshold, wherein the at least one to-be-matched point includes a first to-be-matched point; Obtaining, according to the first to-be-matched point, position coordinates of elements of a distance value matrix corresponding to the first to-be-matched point; Obtaining at least one to-be-matched point set of the second point cloud frame according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched points and the multiple point sets of the second point cloud frame; Calculating a first absolute value, wherein the first absolute value is the absolute value of the difference between the value of the feature parameter of the current set of points to be matched in the second point cloud frame and the value of the element corresponding to the first set of points to be matched; When the first absolute value is less than a third preset threshold, determining whether the first to-be-matched point is a noise point according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched point, the values of the characteristic parameters of the current to-be-matched point set, and the values of the characteristic parameters of the multiple point sets of the first point cloud frame; Target recognition is performed based on the noise point judgment result of the current point cloud frame and the first point to be matched.

2. The method according to claim 1, characterized in that The dividing the points of the first point cloud frame and the points of the second point cloud frame according to the first preset threshold to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame includes: According to the distance value matrix corresponding to each of the point cloud frames, obtain the value of the adjacent first element and the value of the second element in the distance value matrix corresponding to each of the point cloud frames; When the absolute value of the difference between the value of the first element and the value of the second element is less than the first preset threshold, the point corresponding to the first element and the point corresponding to the second element are divided into the same point set; or When the absolute value of the difference between the value of the first element and the value of the second element is greater than or equal to the first preset threshold, the point corresponding to the first element and the point corresponding to the second element are divided into different point sets.

3. The method according to claim 1, characterized in that The characteristic parameter is a combination of one or more of the following parameters: mean value, centroid position, fitting normal vector coefficient, flatness ratio or roundness; Accordingly, obtaining the value of the characteristic parameter of each point set according to the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix, and the second distance value matrix includes: Obtaining an average value of each point set according to the values of the elements corresponding to the points included in each point set; or Obtaining the value of the centroid position of each point set according to the average value of the position coordinates of the elements corresponding to the points included in each point set; or Obtaining the value of the fitting normal vector coefficient of each point set according to the value of the element corresponding to the point included in each point set and the position coordinates of the element corresponding to the point included in each point set; or Obtaining a value of the flatness ratio of each point set according to a range of position coordinates of the elements corresponding to the points included in each point set; or The roundness value of each point set is obtained according to the position coordinates of the elements corresponding to the points included in each point set and the centroid position of each point set.

4. The method according to claim 1, wherein The obtaining at least one to-be-matched point from the current point cloud frame according to the first preset threshold and the second preset threshold includes: Obtaining a first number of elements based on the element of the distance value matrix corresponding to a point in the current point cloud frame and a first preset neighborhood, wherein the first number of elements is the number of elements within the first preset neighborhood whose absolute value of the difference between the value of the element of the distance value matrix corresponding to the point in the current point cloud frame is less than a first preset threshold; When the number of the first elements is greater than the second preset threshold, a point in the current point cloud frame is determined as the point to be matched.

5. The method according to claim 1, wherein The obtaining, according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched points and the plurality of point sets of the second point cloud frame, at least one to-be-matched point set of the second point cloud frame comprises: Obtaining, according to the position coordinates of an element of the distance value matrix corresponding to the first point to be matched, the position coordinates of the element corresponding to a point in the second point cloud frame; Obtaining position coordinates of a plurality of elements within the second preset neighborhood according to the position coordinates of the element corresponding to a point in the second point cloud frame and the second preset neighborhood; At least one point set to be matched is obtained based on the position coordinates of the multiple elements within the second preset neighborhood and the multiple point sets of the second point cloud frame, wherein the point set to be matched is the point set including the points corresponding to at least one element within the second preset neighborhood.

6. The method according to claim 1 or 3, characterized in that The determining whether the first to-be-matched point is a noise point according to the position coordinates of the elements of the distance value matrix corresponding to the first to-be-matched point, the values of the characteristic parameters of the current to-be-matched point set, and the values of the characteristic parameters of the multiple point sets of the first point cloud frame includes: Obtaining an inter-frame distance difference according to an absolute value of a difference between the average value of each point set of the first point cloud frame and the average value of the current point set to be matched; Obtaining an inter-frame tilt angle difference according to the value of the centroid position of each point set of the first point cloud frame, the value of the centroid position of the current point set to be matched, and the position coordinates of the element of the distance value matrix corresponding to the first point to be matched; Obtaining an inter-frame normal vector deviation according to the value of the fitted normal vector coefficient of each point set of the first point cloud frame and the value of the fitted normal vector coefficient of the current point set to be matched; Obtaining an inter-frame divergence difference according to the number of points and the average value of each point set of the first point cloud frame and the number of points and the average value of the current point set to be matched; Obtaining an inter-frame flatness ratio difference according to the flatness ratio value of each point set of the first point cloud frame and the flatness ratio value of the current point set to be matched; Obtaining an inter-frame roundness difference according to the roundness value of each point set of the first point cloud frame and the roundness value of the current point set to be matched; When there is no point set among the multiple point sets of the first point cloud frame, so that the inter-frame distance difference is less than the third preset threshold, the inter-frame inclination difference is less than the fourth preset threshold, the inter-frame normal vector deviation is less than the fifth preset threshold, the inter-frame divergence difference is less than the sixth preset threshold, the inter-frame flatness ratio difference is less than the seventh preset threshold, and the inter-frame roundness difference is less than the eighth preset threshold, the first point to be matched is determined to be the noise point.

7. The method according to claim 1 or 3, characterized in that The method further comprises: Calculating a first absolute value corresponding to each of the to-be-matched point sets in the second point cloud frame; When the first absolute value corresponding to each of the to-be-matched point sets is greater than or equal to the third preset threshold, the first to-be-matched point is a noise point.

8. A target recognition device for laser radar point cloud, characterized in that: include: an acquisition module, configured to obtain a first distance value matrix and a second distance value matrix based on a first point cloud frame and a second point cloud frame acquired by the laser radar, wherein an element of each distance value matrix corresponds one-to-one to a point of each point cloud frame, and the first point cloud frame, the second point cloud frame, and the current point cloud frame are point cloud frames acquired sequentially in time sequence; a point cloud processing module, configured to divide the points of the first point cloud frame and the points of the second point cloud frame according to a first preset threshold value, to obtain multiple point sets of the first point cloud frame and multiple point sets of the second point cloud frame; the point cloud processing module is further configured to obtain a value of a characteristic parameter of each point set based on the multiple point sets of the first point cloud frame, the multiple point sets of the second point cloud frame, the first distance value matrix, and the second distance value matrix; the point cloud processing module is further configured to obtain at least one point to be matched from the current point cloud frame based on the first preset threshold value and the second preset threshold value, wherein the at least one point to be matched includes a first point to be matched; the point cloud processing module is further configured to obtain, based on the first point to be matched, the position coordinates of the elements of the distance value matrix corresponding to the first point to be matched; the point cloud processing module is further configured to obtain, based on the position coordinates of the elements of the distance value matrix corresponding to the first point to be matched and the multiple point sets of the second point cloud frame, at least one point set to be matched in the second point cloud frame; a noise point identification module, the noise point identification module being configured to calculate a first absolute value, wherein the first absolute value is the absolute value of the difference between a value of a characteristic parameter of a current set of points to be matched in the second point cloud frame and a value of the element corresponding to the first point to be matched, the noise point identification module being further configured to determine whether the first point to be matched is a noise point based on the position coordinates of an element of a distance value matrix corresponding to the first point to be matched, the value of the characteristic parameter of the current set of points to be matched, and the values of the characteristic parameters of the multiple point sets in the first point cloud frame when the first absolute value is less than a third preset threshold; A target recognition module is used to perform target recognition based on the current point cloud frame and the noise point judgment result of the first point to be matched.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which is used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images