Moving target detection method and system

Abstract

The invention discloses a moving target detection method and system, and the method enables the ground and other backgrounds to be considered separately, provides a simplified ground rapid estimation algorithm, and has certain adaptability to complex terrains such as gentle slopes. Moving target segmentation is completed through two-time difference of a preorder background and a multi-frame accumulation background, most of background points of the accumulation background are rapidly filtered through radius retrieval, more detailed filtering is carried out on the rest of complex point conditions through changes of grid characteristics such as density by the preorder background, and the false alarm rate is reduced; ground fitting, multi-frame accumulated backgrounds and preorder backgrounds are dynamically updated in real time, long-time automation of an invader detection process is guaranteed, and the method is high in environmental adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of moving target detection, and in particular to a moving target detection method and system. Background Art

[0002] Moving target detection is one of the branches of target detection. It focuses more on moving objects or intruding objects in the scene. Although it does not require the identification of all targets in the scene, it reduces the detection requirements, but the demand for real-time performance is greatly increased. In the detection of moving targets based on 3D laser point clouds, one of the main difficulties is the segmentation of fixed background and moving foreground. For example, the presence of some unstable factors in the background, such as ghosts, tree shaking, plant growth, water movement, occlusion, etc., causes mis-segmentation. In order to solve this problem, some methods try to use the pre-scanned background as a reference. Although the occlusion problem is solved, the plant growth and water movement mentioned above are prone to false alarms; in autonomous driving, road test sensors are established as background references, but the requirements for road test laser equipment are high, and the installation scene is limited to outdoor, which increases the cost; artificial intelligence-based methods also detect by marking moving and static targets, and the cost of obtaining samples is high; traditional methods also use Gaussian mixture models to eliminate false alarms such as tree shaking, but the applicable scenarios are narrow and the calculations are complex.

[0003] There is currently no effective solution to the problems of many false positives, high cost and low computational efficiency in moving target detection in existing technologies. Summary of the invention

[0004] The embodiments of the present invention provide a moving target detection method and system to solve the problems of many false detections, high cost and low computational efficiency in moving target detection in the prior art.

[0005] To achieve the above-mentioned purpose, on the one hand, the present invention provides a moving target detection method, which includes: S1, obtaining the original point cloud of the current frame and performing pre-processing to obtain the pre-processed point cloud of the current frame; S2, obtaining the ground grid set corresponding to the current frame, and the maximum height value of each grid, marking all ground points of the pre-processed point cloud of the current frame according to the maximum height value of each grid, removing all ground points of the pre-processed point cloud of the current frame, and obtaining the non-ground point cloud of the current frame; S3, obtaining the cumulative background point cloud and the cumulative background compensation point cloud corresponding to the current frame, using radius retrieval to mark the points in the non-ground point cloud of the current frame that are too close to the cumulative background point cloud or the cumulative background compensation point cloud, and obtaining the background point cloud of the current frame; S4, obtaining the previous background grid set corresponding to the current frame, and the density, reflectivity distribution and point position distribution of each grid, and gridding the non-ground point cloud of the current frame according to the division method of the previous background grid set. Divide to obtain the non-ground point cloud grid set of the current frame; mark the points in the non-ground point cloud of the current frame that may be the background according to the non-ground point cloud grid set of the current frame and the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and add them to the background point cloud of the current frame to update the background point cloud of the current frame; S5, remove the background point cloud of the current frame from the non-ground point cloud of the current frame to obtain the target point cloud of the current frame; cluster the target point cloud of the current frame to obtain each point cluster; for each point cluster, check whether all the background points of the current frame and all the ground points of the current frame are mislabeled based on radius search and reflectivity standard error. If so, cancel the background or ground annotation and add it to the corresponding point cluster and the target point cloud of the current frame; S6, track the target based on the target point cloud of the current frame, and determine whether the target point cloud of the current frame is misdetected. If so, add the corresponding target point to the returned point set.

[0006] Optionally, the acquisition of the ground grid set corresponding to the current frame and the maximum height value of each grid includes: S21, accumulating the pre-processed point cloud with empty current multi-frame targets to obtain the current given point cloud; S22, rasterizing the current given point cloud to obtain multiple grids; calculating the height average and maximum height value of each grid; S23, judging whether each grid is a valid grid; S24, updating the height average and maximum height value of invalid grids according to the height average and maximum height values ​​of all grids; S25, calculating each according to the height average and maximum height value of each grid. The height change of the grid; S26, accumulate the pre-processed point cloud with empty targets in the next multi-frame to obtain the next given point cloud; repeat S22~S25 to calculate the maximum height value and height change of each grid corresponding to the next given point cloud; S27, according to the maximum height value and height change of each grid corresponding to the current given point cloud, and the maximum height value and height change of each grid corresponding to the next given point cloud, update each grid corresponding to the current given point cloud; S28, if the target of the current frame is not empty, the ground grid set corresponding to the current frame is each grid corresponding to the previous given point cloud.

[0007] Optionally, S23 includes: using a consistent segmentation algorithm to fit possible planes in each grid to obtain a fitted plane for each grid; if the ratio of the total number of points on the fitting plane of the current grid to the total number of points in the current grid is less than a preset ratio value, the current grid is invalid; if the Z value standard deviation of all points on the fitting plane of the current grid is greater than a preset Z value standard deviation value, the current grid is invalid; if the reflectivity standard deviation of all points on the fitting plane of the current grid is greater than a preset reflectivity standard deviation value, the current grid is invalid.

[0008] Optionally, S24 includes: finding the nth row with the largest number of valid grids in the preset first M rows of grids, calculating the height average and maximum height value of all grids in the nth row; replacing the height average and maximum height value of all invalid grids in the 0th row with the height average and maximum height value of all grids in the nth row; for invalid grids in other rows, updating the invalid grids using the following update method: replacing the height average of invalid grids in the current row and column with the height average of grids in the previous row and column; calculating the standard deviation of the height average of all grids in the current row, and the height average and maximum height value of valid grids in the current row; if the standard deviation of the height average of all grids in the current row is less than the preset height average standard deviation, replacing the height average and maximum height value of invalid grids in the current row and column with the height average and maximum height value of valid grids in the current row; if the standard deviation of the height average of all grids in the current row is greater than or equal to the preset height average standard deviation, subtracting the height average from the maximum height value of valid grids in the current row and adding the height average of invalid grids in the current row and column after replacement to replace the maximum height value of invalid grids in the current row and column.

[0009] Optionally, the obtaining of the accumulated background point cloud and accumulated background compensation point cloud corresponding to the current frame includes: the accumulated background point cloud and accumulated background compensation point cloud corresponding to all frames are updated in real time using the following method: S31, accumulating the background point cloud with empty initial multi-frame targets as the initial accumulated background point cloud; accumulating the background compensation point cloud with empty initial multi-frame targets as the accumulated background compensation point cloud, updating the initial accumulated background point cloud according to the initial accumulated background point cloud and the accumulated background compensation point cloud to obtain the accumulated background point cloud; clearing the initial accumulated background point cloud and the accumulated background compensation point cloud; S32, when the current multi-frame target is empty and the time since the last update of the accumulated background does not exceed the threshold value T3 When the current multi-frame target is not empty, only the cumulative background compensation point cloud is updated; S33, when the current multi-frame target is not empty, only the cumulative background compensation point cloud is updated; S34, when the current multi-frame target is empty or not empty, but the time from the last cumulative background update exceeds the threshold value T4, the S31 method is used to update the cumulative background point cloud and the cumulative background compensation point cloud; S35, when the current multi-frame target is empty, and the time from the last multi-frame target is not empty does not exceed the threshold value T1, only the cumulative background compensation point cloud is updated; S36, when the current multi-frame target is empty, and the time from the last multi-frame target is not empty exceeds the threshold value T1, the S31 method is used to update the cumulative background point cloud and the cumulative background compensation point cloud.

[0010] Optionally, the method of obtaining the preceding background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid includes: S41, if the target of the initial frame is empty, removing the target point from the pre-processed point cloud of the initial frame to obtain the preceding background point cloud of the initial frame; grid-dividing the preceding background point cloud of the initial frame to obtain the preceding background grid set of the initial frame; and keeping the preceding background grid set of the initial frame as a standby to obtain a standby preceding background grid set; S42, if the target of the current frame is not empty, obtaining the preceding background grid set of the current frame according to the method of S41, and using the standby preceding background grid set of the previous frame as the standby preceding background grid set of the current frame; S43, if the target of the current frame is empty, and the time from the last time the target was not empty exceeds a threshold value T1, obtaining the preceding background grid set of the current frame according to the method of S41, and recording the preceding background grid set of the current frame as a standby to update the standby preceding background grid set. previous background raster set; S44, if the target of the current frame is empty, and the time from the last acquisition of the backup previous background raster set does not exceed the threshold value T2, then the previous background raster set of the current frame is acquired according to the S41 method, and the points in the previous background raster set of the current frame and the backup previous background raster set are compared. If there are no points in the previous background raster set of the current frame or the number of points is less than the points in the backup previous background raster set, the points in the backup previous background raster set are added to the previous background raster set of the current frame; S45, if the target of the current frame is empty, and the time from the last acquisition of the backup previous background raster set exceeds the threshold value T2, then the previous background raster set of the current frame is acquired according to the S41 method, and the backup previous background raster set of the previous frame is used as the backup previous background raster set of the current frame; S46, calculate the density, reflectivity distribution and point position distribution of each grid in the previous background raster set corresponding to each frame.

[0011] On the other hand, the present invention provides a moving target detection system, which includes: a pre-processing unit, which is used to obtain the original point cloud of the current frame and perform pre-processing to obtain the pre-processed point cloud of the current frame; a ground point removal unit, which is used to obtain the ground grid set corresponding to the current frame, and the maximum height value of each grid, mark all ground points of the pre-processed point cloud of the current frame according to the maximum height value of each grid, remove all ground points of the pre-processed point cloud of the current frame, and obtain the non-ground point cloud of the current frame; a background point cloud acquisition unit, which is used to obtain the cumulative background point cloud and the cumulative background compensation point cloud corresponding to the current frame, use radius retrieval to mark the points in the non-ground point cloud of the current frame that are too close to the cumulative background point cloud or the cumulative background compensation point cloud, and obtain the background point cloud of the current frame; a background point cloud updating unit, which is used to obtain the previous background grid set corresponding to the current frame, and the density, reflectivity distribution and point position distribution of each grid, and update the non-ground point cloud of the current frame according to the division method of the previous background grid set. The grid is divided into rows to obtain the non-ground point cloud grid set of the current frame; the points in the non-ground point cloud of the current frame that may be the background are marked according to the non-ground point cloud grid set of the current frame and the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and added to the background point cloud of the current frame to update the background point cloud of the current frame; the target point cloud acquisition unit is used to remove the background point cloud of the current frame from the non-ground point cloud of the current frame to obtain the target point cloud of the current frame; the target point cloud of the current frame is clustered to obtain each point cluster; for each point cluster, based on radius search and reflectivity standard error, all background points of the current frame and all ground points of the current frame are checked to see if they are mislabeled. If so, the background or ground annotation is cancelled and added to the corresponding point cluster and the target point cloud of the current frame; the target point cloud update unit is used to track the target point cloud of the current frame and determine whether the target point cloud of the current frame is misdetected. If so, the corresponding target point is added to the returned point set.

[0012] Optionally, the method of obtaining the ground grid set corresponding to the current frame and the maximum height value of each grid includes: a current accumulation subunit, which is used to accumulate the pre-processed point cloud in which the current multi-frame target is empty to obtain the current given point cloud; a rasterization subunit, which is used to rasterize the current given point cloud to obtain multiple grids; calculate the height average and maximum height value of each grid; a first judgment subunit, which is used to judge whether each grid is a valid grid; a first update subunit, which is used to update the height average and maximum height value of invalid grids according to the height average and maximum height values ​​of all grids; a first calculation subunit, which is used to calculate the height change of each grid according to the height average and maximum height value of each grid The next accumulation subunit is used to accumulate the pre-processed point cloud with empty targets in the next multiple frames to obtain the next given point cloud; the rasterization subunit, the first judgment subunit, the first update subunit and the first calculation subunit are repeated to calculate the maximum height value and the height change of each grid corresponding to the next given point cloud; the second update subunit is used to update each grid corresponding to the current given point cloud according to the maximum height value and the height change of each grid corresponding to the current given point cloud and the maximum height value and the height change of each grid corresponding to the next given point cloud; the ground grid set acquisition unit is used to, if the target of the current frame is not empty, then the ground grid set corresponding to the current frame is each grid corresponding to the last given point cloud.

[0013] Optionally, the acquisition of the cumulative background point cloud and the cumulative background compensation point cloud corresponding to the current frame includes: the cumulative background point clouds and the cumulative background compensation point clouds corresponding to all frames are updated in real time using the following method: an initial accumulation subunit, used to accumulate the background point clouds with empty initial multi-frame targets as the initial cumulative background point clouds; accumulate the background compensation point clouds with empty initial multi-frame targets as the cumulative background compensation point clouds, update the initial cumulative background point clouds according to the initial cumulative background point clouds and the cumulative background compensation point clouds to obtain the cumulative background point clouds; clear the initial cumulative background point clouds and the cumulative background compensation point clouds; a second judgment subunit, used to only update the cumulative background compensation point clouds when the current multi-frame targets are empty and the time from the last update of the cumulative background does not exceed a threshold value T3 compensation point cloud; a third judgment subunit, used to update only the accumulated background compensation point cloud when the current multi-frame target is not empty; a fourth judgment subunit, used to update the accumulated background point cloud and the accumulated background compensation point cloud according to the initial accumulation subunit method when the current multi-frame target is empty or not empty, but the time from the last cumulative background update exceeds the threshold value T4; a fifth judgment subunit, used to update only the accumulated background compensation point cloud when the current multi-frame target is empty and the time from the last multi-frame target is not empty does not exceed the threshold value T1; a sixth judgment subunit, used to update the accumulated background point cloud and the accumulated background compensation point cloud according to the initial accumulation subunit method when the current multi-frame target is empty and the time from the last multi-frame target is not empty exceeds the threshold value T1.

[0014] Optionally, the method of obtaining the preceding background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid includes: an initial grid division subunit, which is used to remove the target points from the pre-processed point cloud of the initial frame to obtain the preceding background point cloud of the initial frame if the target of the initial frame is empty; grid division of the preceding background point cloud of the initial frame to obtain the preceding background grid set of the initial frame; and setting the preceding background grid set of the initial frame aside to obtain a spare preceding background grid set; a first acquisition subunit, which is used to obtain the preceding background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is not empty, and use the spare preceding background grid set of the previous frame as the spare preceding background grid set of the current frame; a second acquisition subunit, which is used to obtain the preceding background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time from the last time the target was not empty exceeds a threshold value T1, and record the preceding background grid set of the current frame as a spare to update the spare a preceding background grid set; a third acquisition subunit, for acquiring the preceding background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time from the last acquisition of the backup preceding background grid set does not exceed the threshold value T2, and compares the points in the preceding background grid set of the current frame and the backup preceding background grid set, if there are no points in the preceding background grid set of the current frame or the number of points is less than the points in the backup preceding background grid set, then the points in the backup preceding background grid set are added to the preceding background grid set of the current frame; a fourth acquisition subunit, for acquiring the preceding background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time from the last acquisition of the backup preceding background grid set exceeds the threshold value T2, and the backup preceding background grid set of the previous frame is used as the backup preceding background grid set of the current frame; a second calculation subunit, for calculating the density, reflectivity distribution and point position distribution of each grid in the preceding background grid set corresponding to each frame.

[0015] Beneficial effects of the present invention: The present invention provides a moving target detection method and system, wherein the method considers the ground and the rest of the background separately, and provides a simplified algorithm for rapid ground estimation, which has certain adaptability to relatively complex terrains such as gentle slopes. The moving target segmentation is completed by two differences between the preceding background and the multi-frame accumulated background, wherein the accumulated background quickly filters out most of the background points through radius retrieval, and the remaining complex points are filtered more carefully by the preceding background through changes in grid features such as density, thereby reducing the false alarm rate; the ground fitting, multi-frame accumulated background, and preceding background are dynamically updated in real time to ensure long-term automation of the intruder detection process, and have strong adaptability to the environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a moving target detection method provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a moving target detection system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Figure 1 is a flow chart of a moving target detection method provided by an embodiment of the present invention. Figure 1 As shown, the method includes: S1, obtain the original point cloud of the current frame and perform pre-processing to obtain the pre-processed point cloud of the current frame; In an optional implementation, the original point cloud of the current frame is firstly filtered according to the security area, and the effective defense area is cut out to obtain the pre-processed result, which is recorded as the pre-processed point cloud P of the current frame. 0 .

[0019] S2, obtaining the ground grid set corresponding to the current frame and the maximum height value of each grid, marking all ground points of the pre-processed point cloud of the current frame according to the maximum height value of each grid, removing all ground points of the pre-processed point cloud of the current frame, and obtaining the non-ground point cloud of the current frame; Since ground data is greatly affected by the laser incident angle, material, etc., and smooth roads in rainy weather will produce puddle reflections, non-smooth roads will produce splash noise, and the ground itself has obvious features, the ground is segmented separately.

[0020] Get the ground grid set corresponding to the current frame and the maximum height value Zmax of each grid ij (i represents the division on the x-axis, j represents the division on the y-axis); the ground grid set is a two-dimensional grid set; if a point P in the pre-processed point cloud of the current frame 01 (X 0 ,Y 0 ,Z 0 ), one of the grids in the ground grid set corresponding to the current frame is x=0 to 0.5 and y=0 to 0.5, then as long as X 0 ∈(0,0.5) and Y 0 ∈(0,0.5), then this point falls in this grid, and then compare Z 0 and the Zmax of the grid 11 (maximum height value); if Z 0Greater than the Zmax of the grid 11 +Zoffset (given offset), then keep the point, otherwise mark the point as a ground point; All ground points are obtained by the above method, and all ground points of the pre-processed point cloud of the current frame are removed to obtain the non-ground point cloud P of the current frame. 1 .

[0021] In an optional implementation, the step of obtaining a ground grid set corresponding to the current frame and a maximum height value of each grid includes: S21, accumulating the pre-processed point cloud with empty target in the current multi-frame to obtain the current given point cloud; Assuming that the targets are all empty in the initial 0-5 frames, the pre-processed point clouds of these 5 frames are accumulated to obtain the current given point cloud P C0 .

[0022] S22, rasterizing the current given point cloud to obtain multiple grids; calculating the average height and maximum height of each grid; Split the current given point cloud into multiple grids by grid ij , calculate the average height value (that is, the Z mean value of all points in the grid) and the maximum height value (that is, the maximum Z value of all points in the grid) of each grid.

[0023] S23, judging whether each grid is a valid grid; In an optional implementation, the S23 includes: The consistent segmentation algorithm is used to fit the possible planes in each grid to obtain the fitting plane of each grid; If the ratio of the total number of points on the fitted plane of the current grid to the total number of points in the current grid is less than the preset ratio value, the current grid is invalid; If the Z value standard deviation of all points on the fitting plane of the current grid is greater than the preset Z value standard deviation, the current grid is invalid; If the reflectance standard deviation of all points on the fitting plane of the current grid is greater than the preset reflectance standard deviation value, the current grid is invalid.

[0024] Further, the normal vector of the fitting plane of each grid is calculated; For each row of the grid, starting from y=0, calculate to the left and right: 1. The normal vector change rate between adjacent grids (i.e., the normal vector change rate between two adjacent grids). The change rate of an invalid grid relative to a valid grid is recorded as 0. If the change rate is lower than the preset change rate threshold, the adjacent grids are considered to be in the same plane.

[0025] 2. If the rate of change is higher than the preset rate of change threshold, it is considered to be the beginning of a new plane and attempts are made to grow from this grid. If several valid grids in the same plane cannot be found within a finite interval, the grid is marked as an invalid grid.

[0026] For each column of the grid, start from x=0 and proceed in the same way.

[0027] Furthermore, Kmeans clustering is used with K=2 to cluster the valid grids into two parts. The Z-mean of all grids in each part is calculated. If the difference between the Z-means of the two parts is greater than the preset threshold, all grids in the part with the larger Z-mean are considered to be the ceiling, and the grids in this part are recorded as invalid grids. Otherwise, the grids in both parts are considered to be valid grids.

[0028] S24, updating the average height value and the maximum height value of the invalid grids according to the average height value and the maximum height value of all grids; In an optional implementation, the S24 includes: Among the preset first M rows of grids, find the nth row with the largest number of valid grids and calculate the average height Zavg of all grids in the nth row n and the maximum height Zmax n ; Replace the height average and maximum height values ​​of all grids in row n with the height average of all invalid grids in row 0, Zavg 0j and the maximum height Zmax 0j ; For invalid grids in other rows, use the following update method to update the invalid grids: Replace the average height of the invalid grid in the current row and column with the average height of the grid in the previous row and column; That is, for an invalid grid (i, j), use the Zavg of the grid (i-1, j) i-1j Fill the invalid grid (i,j).

[0029] Calculate the standard deviation of the height averages of all grids in the current row, as well as the height average Zavg of the valid grids in the current row. i and the maximum height Zmax i ; If the standard deviation of the height averages of all grids in the current row is less than the preset height average standard deviation, the height average and maximum height values ​​of the valid grids in the current row will be used to replace the height average and maximum height values ​​of the invalid grids in the current row and column. That is, update the Zavg of the invalid grid in the row ij =Zavg i , Zmaxij=Zmaxi; If the standard deviation of the height averages of all grids in the current row is greater than or equal to the preset height average standard deviation, the maximum height value of the valid grid in the current row is subtracted from the height average and then added to the height average of the replaced invalid grid in the current row and column to replace the maximum height value of the invalid grid in the current row and column.

[0030] That is, update the Zmax of the invalid grid in the row ij =Zavg ij (i.e. Zavg i-1j )+(Zmax i -Zavg i ).

[0031] S25, calculating the height variation of each grid according to the height average value and the maximum height value of each grid; Each grid has its own maximum height Zmax ij and height change Zdis ij =Zmax ij -Zavg ij .

[0032] S26, accumulating the pre-processed point cloud with the next multi-frame target being empty to obtain the next given point cloud; repeating S22 to S25, calculating the maximum height value and height change of each grid corresponding to the next given point cloud; In an optional implementation, the next multiple frames where the target is empty are frames 8-13, and the same method as above is used to calculate the ground grid set corresponding to frames 8-13, as well as the maximum height value and height change of each grid.

[0033] S27, updating each grid corresponding to the current given point cloud according to the maximum height value and height change of each grid corresponding to the current given point cloud, and the maximum height value and height change of each grid corresponding to the next given point cloud; In an optional implementation, the ground grid set G corresponding to frames 0-5 ground and the ground grid set G' corresponding to frames 8-13 ground , whether to update G ground , judged as follows: (1) Calculate the difference between the Zmax values ​​of the two grids. If the difference is greater than the height change Zdis ij If the ratio is within the preset range, the two grids are considered similar, otherwise the two grids are different.

[0034] For example: G ground The maximum height value of the first grid and G' ground The difference between the maximum height value of the first grid and G ground The height change of the first grid and G'ground If the ratio of the height change of the first grid is within the preset ratio range, it is considered that G ground The first grid and G' ground The first grid is close to .

[0035] (2) For each column of grids, if a number of different grids are found in a finite interval starting from the nth grid, use G' from the nth grid onwards ground The values ​​(maximum height value, average height value, height change) are updated to G ground .

[0036] S28. If the target of the current frame is not empty, the ground grid set corresponding to the current frame is each grid corresponding to the last given point cloud.

[0037] In an optional implementation, if the targets of the 6th and 7th frames are not empty, the corresponding ground grid sets of the 6th and 7th frames are the ground grid sets corresponding to the given point clouds calculated in frames 0-5, which is equivalent to that the ground grid sets are not updated.

[0038] If the target in frames 8-13 is empty, the ground grid set is updated. If the current frame obtained is a frame within frames 8-13, the ground grid set corresponding to the given point cloud calculated in frames 8-13 is used; If the targets of frames 14 and 15 are not empty, the corresponding ground grid sets of frames 14 and 15 are the ground grid sets corresponding to the given point clouds calculated by frames 8-13.

[0039] S3, obtaining the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame, using the radius to search and mark the points in the non-ground point cloud of the current frame that are too close to the accumulated background point cloud or the accumulated background compensation point cloud, and obtaining the background point cloud of the current frame; In an optional implementation, the accumulated background point cloud P corresponding to the current frame is obtained. cback and the accumulated background compensation point cloud P cpack , compare P cback and P 1 , use the radius to retrieve the marker P 1 Middle distance P cback Too close inside, The specific method is as follows: If P 1 A point in cback If a point in the P is close to each other (i.e., within the radius threshold), then P 1 A point in the P is a background point; the “radius threshold” is not a fixed value, but varies with P. 1 The distance (Dis) from a point in to the origin of the laser radar changes: r back=k×Dis+b Among them, r back is the radius threshold, k and b are given values, Dis is P 1 The distance from a point in to the origin of the laser radar; Assume P 1 The distance Dis from a point A to the origin of the laser radar is 10m, k=0.05 (unit: proportional coefficient), b=0.1 (unit: meter), then the radius threshold of point A is calculated as: r back =0.05×10+0.1=0.6m, that is: if the distance between point A and the accumulated background point cloud P cback If a point in is less than 0.6m, then point A may be a background point and is marked as a background point.

[0040] The same method is used to compare P cpack and P 1 , use the radius to retrieve the marker P 1 Middle distance P cpack The point that is too close inside; By using the above method, all background points in the non-ground point cloud of the current frame are marked to obtain the background point cloud P of the current frame. back .

[0041] In an optional implementation, the acquiring the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame includes: The accumulated background point cloud and accumulated background compensation point cloud corresponding to all frames are updated in real time using the following method: S31, accumulating the background point cloud with empty initial multi-frame targets as the initial accumulated background point cloud; accumulating the background compensation point cloud with empty initial multi-frame targets as the accumulated background compensation point cloud, updating the initial accumulated background point cloud according to the initial accumulated background point cloud and the accumulated background compensation point cloud to obtain the accumulated background point cloud; clearing the initial accumulated background point cloud and the accumulated background compensation point cloud; S32, when the current multi-frame target is empty and the time since the last cumulative background update does not exceed the threshold T3, only the cumulative background compensation point cloud is updated; S33, when the current multi-frame target is not empty, only the accumulated background compensation point cloud is updated; S34, when the current multi-frame target is empty or not empty, but the time since the last cumulative background update exceeds the threshold value T4, the S31 method is used to update the cumulative background point cloud and the cumulative background compensation point cloud; S35, when the current multi-frame target is empty and the time from the last multi-frame target being not empty does not exceed the threshold value T1, only the accumulated background compensation point cloud is updated; S36: When the current multi-frame target is empty and the time from the last time the multi-frame target was not empty exceeds the threshold T1, the S31 method is used to update the accumulated background point cloud and the accumulated background compensation point cloud.

[0042] The following is an explanation using a specific example: Frames 0-5: Initial target P obj and the initial return point P re Empty, the initial background point cloud of 5 frames is used as the initial accumulated background point cloud P' cback The initial return point of each frame is added to the initial background compensation point cloud of each frame, and the initial background compensation point cloud of 5 frames is accumulated as the cumulative background compensation point cloud P cpack ; At the 5th frame, the initial accumulated background point cloud is updated to obtain the accumulated background point cloud P cback =P' cback ∪P cpack , clear P cpack and P' cback ; Frames 6-7: Assume that the target P obj Empty, T3=2, only update the accumulated background compensation point cloud P cpack , which is equivalent to accumulating 6-7 frames of background compensation point cloud to obtain the accumulated background compensation point cloud P cpack , where the accumulated background point cloud is still the accumulated background point cloud P calculated from 0-5 frames cback ; Frames 8-12: Assume that the target P obj Not empty (e.g. 8 is empty, 9 is empty, 10 is not empty, 11 is empty, 12 is not empty), only the accumulated background compensation point cloud P is updated cpack , that is, accumulate the background compensation point cloud of 8-12 frames; the accumulated background point cloud is still the accumulated background point cloud P calculated from 0-5 frames cback ; Frames 13-14: Assume that target P obj Empty, T1=2, only update the accumulated background compensation point cloud P cpack , that is, the background compensation point cloud of 13-14 frames is accumulated again; the accumulated background point cloud is still the accumulated background point cloud P calculated from 0-5 frames cback ; Frames 15-20: Assume that the target P obj Empty, update the accumulated background point cloud P cback , that is, the initial background point cloud of 15-20 frames is taken as the initial accumulated background point cloud P' cback . Update the accumulated background compensation point cloud P cpack , that is, accumulate the background compensation point cloud of 15-20 frames again; update the initial accumulated background point cloud at the 20th frame to obtain the accumulated background point cloud P cback =P' cback ∪P cpack , clear P cpack and P' cback ; 20-120 frames, assuming the target P obj Not empty, only update the accumulated background compensation point cloud P cpack , that is, the background compensation point cloud of the accumulated 20-120 frames; T4=100, where the accumulated background point cloud is still the accumulated background point cloud P calculated for the 15-20 frames cback ; 120-125, regardless of target P obj Is it empty? Update the accumulated background point cloud P cback , that is, the initial background point cloud of 120-125 frames is taken as the initial accumulated background point cloud P' cback . Update the accumulated background compensation point cloud P cpack , that is, accumulate the background compensation point cloud of 120-125 frames again; update the initial accumulated background point cloud at the 125th frame to obtain the accumulated background point cloud P cback =P' cback ∪P cpack , clear P cpack and P' cback .

[0043] S4, obtaining the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and performing grid division on the non-ground point cloud of the current frame according to the division method of the previous background grid set to obtain the non-ground point cloud grid set of the current frame; marking the points that may be background in the non-ground point cloud of the current frame according to the non-ground point cloud grid set of the current frame and the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and adding them to the background point cloud of the current frame to update the background point cloud of the current frame; Get the previous background grid set Grid corresponding to the current frame ij ∈Gpre, as well as the density, reflectivity distribution and point position distribution of each grid; divide the non-ground point cloud P of the current frame according to the same grid division method 1 Get Grid' ij ∈Gcurr, compare Grid ij and Grid' ij (i represents the division on the x-axis, starting from 0, j represents the division on the y-axis, with 0 as the origin, positive on the left and negative on the right).

[0044] (1) If Grid' ij All points in belong to P back , which is marked as background, adding P back ,Finish; (2) If Grid ij There is no point in the middle, it ends; (3) Calculation Grid' ij Are the non-background points in the Gridij If the midpoint position and reflectivity are within the standard error, the non-background points in Grid'ij are marked as background and P is added. back ; (4) If Grid' ij The density of non-background points in is much smaller than that in Grid ij , then Grid' ij All unmarked ones in P are also marked as background and added to P back ; (5) If Grid' ij Non-background point density and Grid in ij If the points are similar, the unmarked points that may be background points are also marked as background points and added to P back ,Finish; (6) Get Grid ij Data Grid in the eight nearby grids IJ To expand the sample, calculate Grid' ij Non-background points in the grid IJ The Wasserstein distance W of the points in IJ (Take four dimensions of data, x, y, z, ref, each dimension has a given weight), in Grid' ij Add points that may be background but are not marked as background one by one, and calculate the Grid' ij To Grid IJ The Wasserstein distance W' of the points in IJ , compare the relative rate of change If the change is large, the point is marked as background and P is added. back .

[0045] Through the above method, the points that may be background in the non-ground point cloud of the current frame are added to the background point cloud P of the current frame. back To update the background point cloud P of the current frame back .

[0046] In an optional implementation, the obtaining of the preceding background grid set corresponding to the current frame, and the density, reflectivity distribution, and point position distribution of each grid includes: S41, if the target of the initial frame is empty, remove the target point from the pre-processed point cloud of the initial frame to obtain a previous background point cloud of the initial frame; grid-divide the previous background point cloud of the initial frame to obtain a previous background grid set of the initial frame; and reserve the previous background grid set of the initial frame to obtain a reserve previous background grid set; S42, if the target of the current frame is not empty, then obtain the previous background grid set of the current frame according to the method of S41, and use the backup previous background grid set of the previous frame as the backup previous background grid set of the current frame; S43, if the target of the current frame is empty, and the time since the last target was not empty exceeds the threshold value T1, then the previous background grid set of the current frame is obtained according to the method of S41, and the previous background grid set of the current frame is recorded for backup to update the backup previous background grid set; S44, if the target of the current frame is empty, and the time since the last acquisition of the backup preceding background grid set does not exceed the threshold value T2, then the preceding background grid set of the current frame is acquired according to the S41 method, and the points in the preceding background grid set of the current frame and the backup preceding background grid set are compared. If there are no points in the preceding background grid set of the current frame or the number of points is less than the points in the backup preceding background grid set, then the points in the backup preceding background grid set are added to the preceding background grid set of the current frame; S45, if the target of the current frame is empty, and the time since the last acquisition of the backup preceding background grid set exceeds the threshold value T2, the preceding background grid set of the current frame is acquired according to the method of S41, and the backup preceding background grid set of the previous frame is used as the backup preceding background grid set of the current frame; S46, calculating the density, reflectivity distribution and point position distribution of each grid in the preceding background grid set corresponding to each frame.

[0047] The following is described by a specific embodiment: Assume T1=2, T2=9.

[0048] Frame 0: If the target is empty, the previous background grid set is G0 (i.e., the previous background point cloud of frame 0 is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G0; Frame 1: If the target is not empty, the previous background grid set is G1 (i.e., the previous background point cloud of frame 1 is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G0; Frame 2: If the target is not empty, the previous background grid set is G2 (i.e., the previous background point cloud of frame 2 is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G0; Frame 3: If the target is empty, the previous background grid set is G3 (that is, the previous background point cloud of frame 3 is grid-divided to obtain the previous background grid set) + the backup previous background grid set is G0, and the backup previous background grid set is G0; (not exceeding T2) Frame 4: If the target is empty, the previous background grid set is G4 (i.e., the previous background point cloud of frame 4 is grid-divided to obtain the previous background grid set) + the backup previous background grid set is G0, and the backup previous background grid set is G0; (not exceeding T2) Frame 5: If the target is empty, the previous background grid set is G5 (i.e., the previous background point cloud of the 5th frame is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G5; (exceeds T1) Frames 6-16: If the target is not empty, the previous background grid set is Gi (i.e., the previous background point cloud of the i-th frame is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G5; Frame 17: If the target is empty, the previous background grid set is G17 (i.e., the previous background point cloud of frame 17 is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G5; (exceeds T2) Frame 18: If the target is empty, the previous background grid set is G18 (i.e., the previous background point cloud of frame 18 is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G5; (exceeds T2) Frame 19: If the target is empty, the previous background grid set is G19 (i.e., the previous background point cloud of frame 19 is grid-divided to obtain the previous background grid set), and the backup previous background grid set is G19; (exceeds T1) S5, removing the background point cloud of the current frame from the non-ground point cloud of the current frame to obtain the target point cloud of the current frame; clustering the target point cloud of the current frame to obtain each point cluster; for each point cluster, checking whether all the background points of the current frame and all the ground points of the current frame are mislabeled based on radius search and reflectivity standard error, if so, canceling the background or ground labeling and adding it to the corresponding point cluster and the target point cloud of the current frame; In an optional embodiment, the background point cloud of the current frame is removed from the non-ground point cloud of the current frame to obtain the target point cloud P of the current frame. obj ; For the target point cloud P of the current frame obj Perform clustering to obtain each point cluster; perform the following operations on each point cluster: (1) Get the specified radius r around the cluster inner point back (The calculation method is consistent with that mentioned in S3) i ∈P back and p j ∈P obj , if p i The reflectivity at p j If the reflectivity of the set is within the standard error, the background label of the point is cancelled and added to the corresponding cluster point set and P obj .

[0049] Set a search radius r back :r back= k×Dis+ b; Dis is the distance from a point in the current cluster to the origin of the laser radar; k and b are given parameters, and the background point p is searched within this radius i ∈P back , if p i The reflectivity at p j If the reflectivity of the set is within the standard error, p is cancelled. i The background annotation of p i Add to the current cluster point cluster and P obj middle.

[0050] (2) Get the specified radius R around the cluster inner point ground (specified value) of point p k ∈P ground and p j ∈P obj , if p k The reflectivity at p j If the reflectivity of the point is within the standard error of the set, the ground annotation of the point is cancelled and added to the corresponding cluster point set and P obj .

[0051] S6. Track the target based on the target point cloud of the current frame, and determine whether the target point cloud of the current frame is misdetected. If so, add the corresponding target point to the returned point set.

[0052] In an optional embodiment, according to P obj The target motion is detected and tracked based on the clustering results. If the target has no obvious displacement between several frames, it is considered to be a newly added background object and added to the return point set P. re If the target appears and disappears occasionally between several frames, it is considered to be a false detection and is added to the return point set P re ; If the target does not appear again after a short period of time, it is considered a false detection and is added to the return point set P re .

[0053] The method of the present invention considers the ground and the rest of the background separately, and provides a simplified algorithm for rapid ground estimation, which has a certain adaptability to relatively complex terrains such as gentle slopes, thereby improving the calculation efficiency; the moving target segmentation is completed by twice differencing the previous background and the multi-frame accumulated background, wherein the accumulated background quickly filters out most of the background points through radius retrieval, and the remaining complex points are filtered more carefully by the previous background through changes in grid features such as density, thereby reducing the false alarm rate; since multiple segmentations may additionally clean up target points, these points are attempted to be returned to the detected target through possible target screening to restore the target shape; the ground fitting, multi-frame accumulated background, and the previous background are dynamically updated in real time to ensure the long-term automation of the intruder detection process, and the method has strong adaptability to the environment.

[0054] Figure 2 is a schematic diagram of the structure of a moving target detection system provided by an embodiment of the present invention, such as Figure 2 As shown, the system includes: The pre-processing unit 201 is used to obtain the original point cloud of the current frame and perform pre-processing to obtain the pre-processed point cloud of the current frame; The ground point removal unit 202 is used to obtain the ground grid set corresponding to the current frame and the maximum height value of each grid, mark all ground points of the pre-processed point cloud of the current frame according to the maximum height value of each grid, remove all ground points of the pre-processed point cloud of the current frame, and obtain the non-ground point cloud of the current frame; The background point cloud acquisition unit 203 is used to acquire the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame, and use the radius to search and mark the points in the non-ground point cloud of the current frame that are too close to the accumulated background point cloud or the accumulated background compensation point cloud to obtain the background point cloud of the current frame; The background point cloud updating unit 204 is used to obtain the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and to grid-divide the non-ground point cloud of the current frame according to the division method of the previous background grid set to obtain the non-ground point cloud grid set of the current frame; mark the points that may be background in the non-ground point cloud of the current frame according to the non-ground point cloud grid set of the current frame and the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and add them to the background point cloud of the current frame to update the background point cloud of the current frame; The target point cloud acquisition unit 205 is used to remove the background point cloud of the current frame from the non-ground point cloud of the current frame to obtain the target point cloud of the current frame; cluster the target point cloud of the current frame to obtain each point cluster; check whether all the background points of the current frame and all the ground points of the current frame are mislabeled based on radius search and reflectivity standard error for each point cluster, and if so, cancel the background or ground labeling and add it to the corresponding point cluster and the target point cloud of the current frame; The target point cloud updating unit 206 is used to track the target point cloud of the current frame and determine whether the target point cloud of the current frame is misdetected. If so, the corresponding target point is added to the return point set.

[0055] In an optional implementation, the step of obtaining a ground grid set corresponding to the current frame and a maximum height value of each grid includes: The current accumulation subunit 2021 is used to accumulate the pre-processed point clouds of the current multi-frame targets being empty to obtain the current given point cloud; The rasterization subunit 2022 is used to rasterize the current given point cloud to obtain multiple grids; calculate the average height and maximum height value of each grid; The first judging subunit 2023 is used to judge whether each grid is a valid grid; A first updating subunit 2024 is used to update the average height value and the maximum height value of the invalid grid according to the average height value and the maximum height value of all grids; The first calculation subunit 2025 is used to calculate the height change of each grid according to the average height value and the maximum height value of each grid; The next accumulation subunit 2026 is used to accumulate the pre-processed point cloud with empty targets in the next multiple frames to obtain the next given point cloud; repeat the rasterization subunit 2022, the first judgment subunit 2023, the first update subunit 2024, and the first calculation subunit 2025 to calculate the maximum height value and height change of each grid corresponding to the next given point cloud; The second updating subunit 2027 is used to update each grid corresponding to the current given point cloud according to the maximum height value and height change amount of each grid corresponding to the current given point cloud, and the maximum height value and height change amount of each grid corresponding to the next given point cloud; The ground grid set acquisition unit 2028 is used to, if the target of the current frame is not empty, then the ground grid set corresponding to the current frame is each grid corresponding to the last given point cloud.

[0056] In an optional implementation, the acquiring the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame includes: The accumulated background point cloud and accumulated background compensation point cloud corresponding to all frames are updated in real time using the following method: The initial accumulation subunit 2031 is used to accumulate the background point cloud with empty initial multi-frame targets as the initial accumulated background point cloud; accumulate the background compensation point cloud with empty initial multi-frame targets as the accumulated background compensation point cloud, update the initial accumulated background point cloud according to the initial accumulated background point cloud and the accumulated background compensation point cloud to obtain the accumulated background point cloud; clear the initial accumulated background point cloud and the accumulated background compensation point cloud; The second judgment subunit 2032 is used to update only the accumulated background compensation point cloud when the current multi-frame target is empty and the time from the last update of the accumulated background does not exceed the threshold value T3; The third judgment subunit 2033 is used to update only the accumulated background compensation point cloud when the current multi-frame target is not empty; The fourth judgment subunit 2034 is used to update the accumulated background point cloud and the accumulated background compensation point cloud according to the initial accumulation subunit method when the current multi-frame target is empty or not empty, but the time since the last cumulative background update exceeds the threshold value T4; The fifth judgment subunit 2035 is used to update only the accumulated background compensation point cloud when the current multi-frame target is empty and the time from the last multi-frame target being not empty does not exceed the threshold value T1; The sixth judgment subunit 2036 is used to update the accumulated background point cloud and the accumulated background compensation point cloud according to the initial accumulation subunit method when the current multi-frame target is empty and the time from the last multi-frame target being not empty exceeds the threshold T1.

[0057] In an optional implementation, the obtaining of the preceding background grid set corresponding to the current frame, and the density, reflectivity distribution, and point position distribution of each grid includes: The initial grid division subunit 2041 is used for removing the target point from the pre-processed point cloud of the initial frame to obtain the previous background point cloud of the initial frame if the target of the initial frame is empty; performing grid division on the previous background point cloud of the initial frame to obtain the previous background grid set of the initial frame; and using the previous background grid set of the initial frame as standby to obtain a standby previous background grid set; The first acquisition subunit 2042 is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is not empty, and use the backup previous background grid set of the previous frame as the backup previous background grid set of the current frame; The second acquisition subunit 2043 is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time since the last target was not empty exceeds the threshold value T1, and record the previous background grid set of the current frame for backup to update the backup previous background grid set; The third acquisition subunit 2044 is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time since the last acquisition of the backup previous background grid set does not exceed the threshold value T2, and compare the points in the previous background grid set of the current frame and the backup previous background grid set. If there are no points in the previous background grid set of the current frame or the number of points is less than the points in the backup previous background grid set, then the points in the backup previous background grid set are added to the previous background grid set of the current frame; The fourth acquisition subunit 2045 is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time from the last acquisition of the backup previous background grid set exceeds the threshold value T2, and use the backup previous background grid set of the previous frame as the backup previous background grid set of the current frame; The second calculation subunit 2046 is used to calculate the density, reflectivity distribution and point position distribution of each grid in the preceding background grid set corresponding to each frame.

[0058] The system of the present invention corresponds to the above-mentioned method, and the specific implementation of the system will not be repeated here.

[0059] Beneficial effects of the present invention: The present invention provides a moving target detection method and system, wherein the method considers the ground and the rest of the background separately, and provides a simplified algorithm for rapid ground estimation, which has certain adaptability to relatively complex terrains such as gentle slopes. The moving target segmentation is completed by two differences between the preceding background and the multi-frame accumulated background, wherein the accumulated background quickly filters out most of the background points through radius retrieval, and the remaining complex points are filtered more carefully by the preceding background through changes in grid features such as density, thereby reducing the false alarm rate; the ground fitting, multi-frame accumulated background, and preceding background are dynamically updated in real time to ensure long-term automation of the intruder detection process, and have strong adaptability to the environment.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A moving target detection method, characterized in that: include: S1, obtain the original point cloud of the current frame and perform pre-processing to obtain the pre-processed point cloud of the current frame; S2, obtaining the ground grid set corresponding to the current frame and the maximum height value of each grid, marking all ground points of the pre-processed point cloud of the current frame according to the maximum height value of each grid, removing all ground points of the pre-processed point cloud of the current frame, and obtaining the non-ground point cloud of the current frame; S3, obtaining the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame, using the radius to search and mark the points in the non-ground point cloud of the current frame that are too close to the accumulated background point cloud or the accumulated background compensation point cloud, and obtaining the background point cloud of the current frame; S4, obtaining the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and performing grid division on the non-ground point cloud of the current frame according to the division method of the previous background grid set to obtain the non-ground point cloud grid set of the current frame; marking the points that may be background in the non-ground point cloud of the current frame according to the non-ground point cloud grid set of the current frame and the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and adding them to the background point cloud of the current frame to update the background point cloud of the current frame; S5, removing the background point cloud of the current frame from the non-ground point cloud of the current frame to obtain the target point cloud of the current frame; clustering the target point cloud of the current frame to obtain each point cluster; for each point cluster, checking whether all the background points of the current frame and all the ground points of the current frame are mislabeled based on radius search and reflectivity standard error, if so, canceling the background or ground labeling and adding it to the corresponding point cluster and the target point cloud of the current frame; S6. Track the target based on the target point cloud of the current frame, and determine whether the target point cloud of the current frame is misdetected. If so, add the corresponding target point to the returned point set.

2. The method according to claim 1, characterized in that The obtaining of the ground grid set corresponding to the current frame and the maximum height value of each grid includes: S21, accumulating the pre-processed point cloud with empty target in the current multi-frame to obtain the current given point cloud; S22, rasterizing the current given point cloud to obtain multiple grids; calculating the average height and maximum height value of each grid; S23, judging whether each grid is a valid grid; S24, updating the average height value and the maximum height value of the invalid grids according to the average height value and the maximum height value of all grids; S25, calculating the height variation of each grid according to the height average value and the maximum height value of each grid; S26, accumulating the pre-processed point cloud with the next multi-frame target being empty to obtain the next given point cloud; repeating S22 to S25, calculating the maximum height value and height change of each grid corresponding to the next given point cloud; S27, updating each grid corresponding to the current given point cloud according to the maximum height value and height change of each grid corresponding to the current given point cloud, and the maximum height value and height change of each grid corresponding to the next given point cloud; S28. If the target of the current frame is not empty, the ground grid set corresponding to the current frame is each grid corresponding to the last given point cloud.

3. The method according to claim 2, characterized in that The S23 includes: The consistent segmentation algorithm is used to fit the possible planes in each grid to obtain the fitting plane of each grid; If the ratio of the total number of points on the fitting plane of the current grid to the total number of points in the current grid is less than the preset ratio value, the current grid is invalid; If the Z value standard deviation of all points on the fitting plane of the current grid is greater than the preset Z value standard deviation, the current grid is invalid; If the reflectance standard deviation of all points on the fitting plane of the current grid is greater than the preset reflectance standard deviation value, the current grid is invalid.

4. The method according to claim 2, characterized in that: The S24 includes: Among the preset first M rows of grids, find the nth row with the largest number of valid grids, calculate the average height and maximum height value of all grids in the nth row; replace the average height and maximum height value of all invalid grids in the 0th row with the average height and maximum height value of all grids in the nth row; For invalid grids in other rows, use the following update method to update the invalid grids: Replace the average height of the invalid grid in the current row and column with the average height of the grid in the previous row and column; Calculate the standard deviation of the height averages of all grids in the current row, as well as the height average and maximum height of the valid grids in the current row; If the standard deviation of the height averages of all grids in the current row is less than the preset height average standard deviation, the height average and maximum height values ​​of the valid grids in the current row will be used to replace the height average and maximum height values ​​of the invalid grids in the current row and column. If the standard deviation of the height averages of all grids in the current row is greater than or equal to the preset height average standard deviation, the maximum height value of the valid grid in the current row is subtracted from the height average and then added to the height average of the replaced invalid grid in the current row and column to replace the maximum height value of the invalid grid in the current row and column.

5. The method according to claim 1, characterized in that The step of obtaining the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame includes: The accumulated background point cloud and accumulated background compensation point cloud corresponding to all frames are updated in real time using the following method: S31, accumulating the background point cloud with empty initial multi-frame targets as the initial accumulated background point cloud; accumulating the background compensation point cloud with empty initial multi-frame targets as the accumulated background compensation point cloud, updating the initial accumulated background point cloud according to the initial accumulated background point cloud and the accumulated background compensation point cloud to obtain the accumulated background point cloud; clearing the initial accumulated background point cloud and the accumulated background compensation point cloud; S32, when the current multi-frame target is empty and the time since the last cumulative background update does not exceed the threshold T3, only the cumulative background compensation point cloud is updated; S33, when the current multi-frame target is not empty, only the accumulated background compensation point cloud is updated; S34, when the current multi-frame target is empty or not empty, but the time since the last cumulative background update exceeds the threshold value T4, the S31 method is used to update the cumulative background point cloud and the cumulative background compensation point cloud; S35, when the current multi-frame target is empty and the time from the last multi-frame target being not empty does not exceed the threshold value T1, only the accumulated background compensation point cloud is updated; S36: When the current multi-frame target is empty and the time from the last time the multi-frame target was not empty exceeds the threshold T1, the S31 method is used to update the accumulated background point cloud and the accumulated background compensation point cloud.

6. The method according to claim 1, characterized in that The obtaining of the preceding background grid set corresponding to the current frame, and the density, reflectivity distribution and point position distribution of each grid comprises: S41, if the target of the initial frame is empty, remove the target point from the pre-processed point cloud of the initial frame to obtain a previous background point cloud of the initial frame; grid-divide the previous background point cloud of the initial frame to obtain a previous background grid set of the initial frame; and reserve the previous background grid set of the initial frame to obtain a reserve previous background grid set; S42, if the target of the current frame is not empty, then obtain the previous background grid set of the current frame according to the method of S41, and use the backup previous background grid set of the previous frame as the backup previous background grid set of the current frame; S43, if the target of the current frame is empty, and the time since the last target was not empty exceeds the threshold value T1, then the previous background grid set of the current frame is obtained according to the method of S41, and the previous background grid set of the current frame is recorded for backup to update the backup previous background grid set; S44, if the target of the current frame is empty, and the time since the last acquisition of the backup preceding background grid set does not exceed the threshold value T2, then the preceding background grid set of the current frame is acquired according to the S41 method, and the points in the preceding background grid set of the current frame and the backup preceding background grid set are compared. If there are no points in the preceding background grid set of the current frame or the number of points is less than the points in the backup preceding background grid set, then the points in the backup preceding background grid set are added to the preceding background grid set of the current frame; S45, if the target of the current frame is empty, and the time since the last acquisition of the backup preceding background grid set exceeds the threshold value T2, the preceding background grid set of the current frame is acquired according to the method of S41, and the backup preceding background grid set of the previous frame is used as the backup preceding background grid set of the current frame; S46, calculating the density, reflectivity distribution and point position distribution of each grid in the preceding background grid set corresponding to each frame.

7. A moving target detection system, characterized in that: include: A pre-processing unit, used for acquiring the original point cloud of the current frame and performing pre-processing to obtain the pre-processed point cloud of the current frame; A ground point removal unit is used to obtain a ground grid set corresponding to the current frame and a maximum height value of each grid, mark all ground points of the pre-processed point cloud of the current frame according to the maximum height value of each grid, remove all ground points of the pre-processed point cloud of the current frame, and obtain a non-ground point cloud of the current frame; A background point cloud acquisition unit is used to acquire the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame, and use the radius to search and mark the points in the non-ground point cloud of the current frame that are too close to the accumulated background point cloud or the accumulated background compensation point cloud to obtain the background point cloud of the current frame; A background point cloud updating unit is used to obtain the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and to grid-divide the non-ground point cloud of the current frame according to the division method of the previous background grid set to obtain the non-ground point cloud grid set of the current frame; and to mark points that may be background in the non-ground point cloud of the current frame according to the non-ground point cloud grid set of the current frame and the previous background grid set corresponding to the current frame, as well as the density, reflectivity distribution and point position distribution of each grid, and add them to the background point cloud of the current frame to update the background point cloud of the current frame; The target point cloud acquisition unit is used to remove the background point cloud of the current frame from the non-ground point cloud of the current frame to obtain the target point cloud of the current frame; cluster the target point cloud of the current frame to obtain each point cluster; for each point cluster, based on radius search and reflectivity standard error, check whether all background points of the current frame and all ground points of the current frame are mislabeled. If so, cancel the background or ground annotation and add it to the corresponding point cluster and the target point cloud of the current frame; The target point cloud updating unit is used to track the target point cloud of the current frame and determine whether the target point cloud of the current frame is misdetected. If so, the corresponding target point is added to the return point set.

8. The system according to claim 7, characterized in that The obtaining of the ground grid set corresponding to the current frame and the maximum height value of each grid includes: The current accumulation subunit is used to accumulate the pre-processed point cloud with empty current multi-frame targets to obtain the current given point cloud; The rasterization subunit is used to rasterize the current given point cloud to obtain multiple grids; calculate the average height and maximum height value of each grid; A first judging subunit, used to judge whether each grid is a valid grid; A first updating subunit, used for updating the average height value and the maximum height value of the invalid grid according to the average height value and the maximum height value of all grids; A first calculation subunit, used for calculating the height variation of each grid according to the average height value and the maximum height value of each grid; The next accumulation subunit is used to accumulate the pre-processed point cloud with empty targets in the next multi-frames to obtain the next given point cloud; repeat the rasterization subunit, the first judgment subunit, the first update subunit, and the first calculation subunit to calculate the maximum height value and height change of each grid corresponding to the next given point cloud; A second updating subunit is used to update each grid corresponding to the current given point cloud according to the maximum height value and height change amount of each grid corresponding to the current given point cloud, and the maximum height value and height change amount of each grid corresponding to the next given point cloud; The ground grid set acquisition unit is used to, if the target of the current frame is not empty, then the ground grid set corresponding to the current frame is each grid corresponding to the last given point cloud.

9. The system according to claim 7, characterized in that The obtaining of the accumulated background point cloud and the accumulated background compensation point cloud corresponding to the current frame comprises: The accumulated background point cloud and accumulated background compensation point cloud corresponding to all frames are updated in real time using the following method: The initial accumulation subunit is used to accumulate the background point cloud with empty initial multi-frame targets as the initial accumulated background point cloud; accumulate the background compensation point cloud with empty initial multi-frame targets as the accumulated background compensation point cloud, update the initial accumulated background point cloud according to the initial accumulated background point cloud and the accumulated background compensation point cloud to obtain the accumulated background point cloud; clear the initial accumulated background point cloud and the accumulated background compensation point cloud; The second judgment subunit is used to update only the accumulated background compensation point cloud when the current multi-frame target is empty and the time from the last cumulative background update does not exceed the threshold value T3; The third judgment subunit is used to update only the accumulated background compensation point cloud when the current multi-frame target is not empty; The fourth judgment subunit is used to update the accumulated background point cloud and the accumulated background compensation point cloud according to the initial accumulation subunit method when the current multi-frame target is empty or not empty, but the accumulated background time since the last update exceeds the threshold value T4; A fifth judgment subunit is used to update only the accumulated background compensation point cloud when the current multi-frame target is empty and the time from the last multi-frame target being not empty does not exceed the threshold value T1; The sixth judgment subunit is used to update the accumulated background point cloud and the accumulated background compensation point cloud according to the initial accumulation subunit method when the current multi-frame target is empty and the time from the last multi-frame target being not empty exceeds the threshold T1.

10. The system according to claim 7, characterized in that The obtaining of the preceding background grid set corresponding to the current frame, and the density, reflectivity distribution and point position distribution of each grid comprises: The initial grid division subunit is used for removing the target point from the pre-processed point cloud of the initial frame to obtain the previous background point cloud of the initial frame if the target of the initial frame is empty; performing grid division on the previous background point cloud of the initial frame to obtain the previous background grid set of the initial frame; and reserving the previous background grid set of the initial frame to obtain a reserve previous background grid set; The first acquisition subunit is used for acquiring the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is not empty, and using the backup previous background grid set of the previous frame as the backup previous background grid set of the current frame; The second acquisition subunit is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time since the last target was not empty exceeds the threshold value T1, and record the previous background grid set of the current frame for backup to update the backup previous background grid set; The third acquisition subunit is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time since the last acquisition of the backup previous background grid set does not exceed the threshold value T2, and compare the points in the previous background grid set of the current frame and the backup previous background grid set. If there are no points in the previous background grid set of the current frame or the number of points is less than the points in the backup previous background grid set, the points in the backup previous background grid set are added to the previous background grid set of the current frame; The fourth acquisition subunit is used to acquire the previous background grid set of the current frame according to the initial grid division subunit method if the target of the current frame is empty and the time from the last acquisition of the backup previous background grid set exceeds the threshold value T2, and use the backup previous background grid set of the previous frame as the backup previous background grid set of the current frame; The second calculation subunit is used to calculate the density, reflectivity distribution and point position distribution of each grid in the preceding background grid set corresponding to each frame.

Images