Automatic acquisition method and system for vehicle point clouds at specific locations and angles

Abstract

This invention discloses an automatic point cloud acquisition method and system for vehicles at specific locations and angles. The method includes: acquiring a frame of point cloud data when there are no vehicles; performing RANSAC fitting on the preprocessed point cloud data to obtain a ground plane; calculating a three-dimensional transformation matrix based on the ground plane and performing three-dimensional calibration on the preprocessed point cloud data; and performing range filtering on the calibrated point cloud to obtain the point cloud PC. masked Then perform RANSAC fitting to obtain the rotation matrix and apply it to the point cloud PC. masked Three-dimensional calibration is performed; histogram statistics are generated from the calibrated point cloud to obtain an overall mask; based on the three-dimensional transformation matrix and rotation matrix, an overall transformation matrix is ​​obtained; based on the overall transformation matrix and overall mask, the real-time acquired point cloud is subjected to three-dimensional calibration and range filtering, followed by 3D mesh segmentation. The timing of point cloud acquisition is selected based on the changes in the centroid spatial coordinates and average reflectivity of the mesh. This invention can adaptively select the timing of acquiring point clouds of approaching vehicles, avoiding the acquisition of empty lane information and duplicate vehicle position information, saving storage space and subsequent data cleaning time.

Description

Technical Field

[0001] This invention relates to the field of point cloud data acquisition technology, and in particular to a method, system, terminal device, and computer-readable storage medium for automatic acquisition of vehicle point clouds at specific locations and angles. Background Technology

[0002] Currently available road point cloud datasets such as KITTI, SynthCity, nuScenes, and Waymo cannot meet the specific needs of certain scenarios. A typical KITTI dataset is collected by multiple sensors on autonomous vehicles, and its height and angle do not match the specific scenario. In particular, there is too little data for various large vehicles and trucks, and the data is unbalanced among different types of vehicles. Most other publicly available datasets are only suitable for autonomous driving and cannot be applied to specific scenario requirements.

[0003] Several typical point cloud data acquisition schemes are as follows: (1) Perform 3D modeling of objects in the scene, and then sample the facets of each vertex to obtain point clouds, such as patent documents CN 112562067A and CN119091079A; (2) For situations where radar scanning targets may have insufficient information, use a preset neural network to enhance the data before sampling, such as patent document CN112396067B; (3) Use multiple acquisition devices to collect local block data of the target from different angles, and then perform point cloud fusion, such as patent documents CN116359942A and CN118898544A. For specific scenario requirements, real-time acquisition by a single radar and a single industrial control device is not suitable for the above schemes in terms of data acquisition hardware and software environment, considering the data acquisition cost and operational complexity.

[0004] In specific scenarios, radar installation is limited by the actual height of the on-site support structure, such as the varying heights of different tollbooths on highways. Furthermore, different technicians employ different installation methods, only ensuring a general radar installation orientation. Inconsistencies in installation position and angle adjustments lead to differences in the coordinate system. Firstly, the position and orientation of the same object in different datasets will be incorrectly labeled due to these differences. Secondly, the model cannot correctly learn spatial relationships during training, causing problems in algorithm evaluation. Simultaneously, ground calibration in the height direction is necessary to reduce data complexity and improve detection accuracy. Generally, manually adjusting parameters after on-site installation is very inefficient.

[0005] In actual data collection, since it's not necessary to collect all data frames sent by the radar when a vehicle approaches, the collection interval is usually set manually. This inevitably leads to the collection of empty lane information (no vehicles, stationary vehicles) and duplicate vehicle positions. Excessive point cloud data not only burdens industrial control computers with insufficient storage space but also significantly increases the workload of data cleaning, reducing the efficiency of data preparation and annotation. Summary of the Invention

[0006] To address at least one of the technical problems in the prior art, this invention provides a method, system, terminal device, and computer-readable storage medium for automatic vehicle point cloud acquisition at specific locations and angles. It automatically obtains approximate equations for the ground surface plane and straight lines on both sides of the lane by fitting and locating the ground detection area based on Random Sample Consistency Multiplexing (RANSAC). It automatically selects the point cloud acquisition timing based on the comprehensive changes in the centroid spatial coordinates and average reflectivity of a 3D segmented grid, thus achieving automatic point cloud data acquisition. Furthermore, it maintains the scale consistency of data acquired by radars installed at different locations, effectively improving debugging and data cleaning efficiency.

[0007] The first objective of this invention is to provide an automatic method for acquiring vehicle point clouds at specific locations and angles.

[0008] The second objective of this invention is to provide an automatic point cloud acquisition system for vehicles at specific locations and angles.

[0009] The third objective of this invention is to provide a terminal device.

[0010] A fourth objective of this invention is to provide a computer-readable storage medium.

[0011] The first objective of this invention can be achieved by adopting the following technical solution:

[0012] An automatic point cloud acquisition method for vehicles at specific locations and angles, the method comprising:

[0013] Collect a frame of point cloud data when there are no cars as the initial point cloud data;

[0014] RANSAC fitting was performed on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set.

[0015] Based on the ground plane, calculate the three-dimensional transformation matrix; use the three-dimensional transformation matrix to perform three-dimensional calibration on the preprocessed initial point cloud data;

[0016] Based on the point cloud set, range filtering is performed on the 3D calibrated point cloud to obtain the point cloud PC. masked For point cloud PCs maskedPerform RANSAC fitting to obtain a straight line in the direction of the road channel;

[0017] Calculate the rotation matrix based on the straight line of the road channel; use the rotation matrix to manipulate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud;

[0018] Histogram statistics are performed on the corrected point cloud to calculate lane positions; based on the lane positions, the overall mask is obtained.

[0019] Based on the three-dimensional transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained;

[0020] The system acquires and preprocesses point clouds in real time; performs 3D calibration and range filtering on the preprocessed point clouds based on the overall transformation matrix and overall mask; performs 3D mesh segmentation on the range-filtered point clouds; and selects the timing of point cloud acquisition based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh to achieve automatic acquisition of vehicle point cloud data.

[0021] Furthermore, before real-time point cloud acquisition, the following steps are also included:

[0022] Point cloud data of n consecutive frames without vehicles were acquired and preprocessed. Then, based on the overall transformation matrix and overall mask, 3D calibration and range filtering were performed on the preprocessed point cloud data. Next, the range-filtered point cloud was divided into 3D meshes along the vehicle's driving direction, and the average z-coordinate z of the point cloud within each 3D mesh of the n frames was calculated. k0 Average x-coordinate k0 and average maximum and minimum reflectivity R kmin R kmax As parameters for car-free lanes; where k represents the grid number, and n is greater than or equal to 2 and less than or equal to ran. x A positive integer of 3, ran x This represents the effective range of the radar scan.

[0023] Furthermore, the real-time acquisition of point clouds is preprocessed; the preprocessed point clouds are then subjected to 3D calibration and range filtering based on the overall transformation matrix and overall mask; the range-filtered point clouds are segmented into 3D meshes, and the timing of point cloud acquisition is selected based on the changes in the centroid spatial coordinates and average reflectivity of the 3D meshes, including:

[0024] A frame of point cloud data is acquired and preprocessed; then, based on the overall transformation matrix and overall mask, 3D calibration and range filtering are performed on the preprocessed point cloud data; finally, the range-filtered point cloud is divided into 3D meshes along the vehicle's driving direction, and the average z-coordinate height z of the point cloud within each 3D mesh is calculated. k Average x-coordinate depth x k and average reflectivity R k ;

[0025] Calculate the Boolean value of the acquisition conditions

[0026] If Collection is 0, then return to collect one frame of point cloud data, perform preprocessing, and continue with subsequent operations; if Collection is 1, let x... k0 =x k Save the current point cloud frame, return to collect a frame of point cloud data, perform preprocessing, and continue with subsequent operations.

[0027] Furthermore, the step of performing histogram statistics on the corrected point cloud and calculating the lane position includes:

[0028] Histogram statistics were performed on the corrected point cloud along the y-axis: a total number of segments (bin) was set, and the coordinate range corresponding to each segmented region was (E... i E i+1 ), count the number of point clouds that fall within this coordinate range;

[0029] Find the index i of the maximum number of point clouds in the histogram statistics. max ;

[0030] If i max If ≤bin / 2, then the y-coordinate of the left side of the lane is E. imax+1 -w, the y-coordinate of the right side of the lane is E imax+1 ; if i max If the value is greater than bin / 2, then the y-coordinate of the left side of the lane is E. imax The y-coordinate of the right side of the lane is E. imax +w; where w is the standard toll station exit width.

[0031] Furthermore, the equation of the surface plane is A0x + B0y + C0z + D0 = 0; where x, y, and z represent x, y, and z coordinates respectively, and A0, B0, C0, and D0 are plane equation parameters.

[0032] The calculation of the three-dimensional transformation matrix based on the ground plane includes:

[0033] The surface plane is transformed into a plane equation of the form z = a0x + a1y + a2, as follows:

[0034]

[0035] Based on the plane equation, we obtain the rotation matrix R and the translation matrix T:

[0036]

[0037] in, h represents the altitude above the ground when the radar acquires the initial point cloud data;

[0038] Based on the rotation matrix R and the translation matrix T, the three-dimensional transformation matrix is ​​obtained.

[0039] Furthermore, the calculation of the rotation matrix based on the straight line of the road channel direction includes:

[0040] Let the projection of the straight line of the road channel onto the xoy plane be y = a3x + a4;

[0041] To transform the line y = a3x + a4 into y = 0, calculate the rotation angle θ = arctan(a3);

[0042] Based on the rotation angle, the corresponding rotation matrix is ​​obtained.

[0043] Furthermore, the step of performing RANSAC fitting on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set includes:

[0044] From the preprocessed initial point cloud data PC dn Three non-collinear points are randomly selected to uniquely determine plane M; the equation of plane M is Ax + By + Cz + D = 0, where x, y, and z represent the x, y, and z coordinates respectively, and A, B, C, and D are parameters of the plane equation.

[0045] Traversing point clouds PC dn Given all points, calculate the distance from any point to plane M:

[0046]

[0047] Statistical satisfaction d m ≤t m The number of points n1; t m The set planar distance threshold;

[0048] After iterating a specified number of times, the plane corresponding to the largest n1 is taken as the surface plane, and the plane with a distance ≤t from the surface plane is considered as the surface plane. m The set of all points is called the point cloud set.

[0049] Furthermore, the point cloud PC masked Perform RANSAC fitting to obtain the straight line in the road channel direction, including:

[0050] From point cloud data PC masked Two points with a z-coordinate difference less than a specified value are randomly selected to determine the line L; the equation of the line L is (x-x0) / a=(y-y0) / b=(z-z0) / c, where x0, y0, z0, a, b, and c are all parameters of the line equation.

[0051] Traversing point clouds PC masked Given all points, calculate the distance from any point to line L:

[0052]

[0053] Statistical satisfaction d l ≤t l The number of points n2; t l The specified straight-line distance threshold;

[0054] Iterate a specified number of times, and take the line corresponding to the maximum n2 as the straight line of the road channel direction.

[0055] Furthermore, the preprocessing includes:

[0056] Set the neighborhood radius ε and the minimum number of points minPTS;

[0057] Traverse all points in the point cloud and calculate the number of points PTS in the ε-neighborhood of each point; if PTS ≥ minPTS, then the point is a core point; otherwise, temporarily mark it as a noise point.

[0058] All core points and all points within their ε-neighborhood are output as denoised point clouds, and the remaining points are removed as noise points. The second objective of this invention can be achieved by adopting the following technical solution:

[0059] An automatic point cloud acquisition system for vehicles at specific locations and angles, the system comprising:

[0060] The initial acquisition module is used to acquire a frame of point cloud data when there are no vehicles as the initial point cloud data.

[0061] The first fitting module is used to perform RANSAC fitting on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set.

[0062] The first calibration module is used to calculate the three-dimensional transformation matrix based on the ground plane; and to perform three-dimensional calibration on the preprocessed initial point cloud data using the three-dimensional transformation matrix.

[0063] The second fitting module is used to perform range filtering on the 3D calibrated point cloud based on the point cloud set to obtain the point cloud PC. masked For point cloud PCs masked Perform RANSAC fitting to obtain a straight line in the direction of the road channel;

[0064] The second calibration module is used to calculate the rotation matrix based on the straight line of the road channel direction; and to use the rotation matrix to calibrate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud;

[0065] The calculation module is used to perform histogram statistics on the corrected point cloud and calculate the lane position; based on the lane position, the overall mask is obtained; and based on the 3D transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained.

[0066] The real-time acquisition module is used to acquire point clouds in real time and perform preprocessing; based on the overall transformation matrix and overall mask, the preprocessed point cloud is then subjected to 3D calibration and range filtering; the range-filtered point cloud is segmented into 3D meshes, and the timing of point cloud acquisition is selected based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh, so as to realize the automatic acquisition of vehicle point cloud data.

[0067] The third objective of this invention can be achieved by adopting the following technical solution:

[0068] A terminal device includes a processor and a memory for storing processor-executable programs. When the processor executes the program stored in the memory, it implements the above-described method for automatically acquiring vehicle point clouds at specific locations and angles.

[0069] The fourth objective of this invention can be achieved by adopting the following technical solution:

[0070] A computer-readable storage medium storing a program that, when executed by a processor, implements the above-described method for automatically acquiring vehicle point clouds at specific locations and angles.

[0071] The present invention has the following advantages over the prior art:

[0072] 1. This invention can automatically locate road surface areas and perform automatic three-dimensional calibration without requiring technicians to manually adjust parameters such as rotation angle and displacement distance in spatial coordinates, which greatly improves work efficiency and ensures detection accuracy.

[0073] 2. This invention can adaptively select the timing for collecting point cloud data of approaching vehicles, avoiding the collection of empty lane information when there are no vehicles or when vehicles are stationary, as well as duplicate vehicle position information; in the case of unmanned operation, it greatly saves equipment storage space and saves subsequent data cleaning time. Attached Figure Description

[0074] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0075] Figure 1This is a simplified flowchart of the automatic vehicle point cloud acquisition method for specific positions and angles according to Embodiment 1 of the present invention.

[0076] Figure 2 This is a detailed flowchart of the automatic vehicle point cloud acquisition method for specific positions and angles according to Embodiment 1 of the present invention.

[0077] Figure 3 This is a diagram illustrating the effect of using RANSAC to generate a fitted land surface plane in Embodiment 1 of the present invention.

[0078] Figure 4 This is a diagram illustrating the effect of using RANSAC to generate a fitted road surface channel in Embodiment 1 of the present invention.

[0079] Figure 5 (a), (b), and (c) are three consecutive point cloud images automatically acquired in Embodiment 1 of the present invention.

[0080] Figure 6 This is a structural block diagram of the automatic vehicle point cloud acquisition system for specific positions and angles according to Embodiment 2 of the present invention.

[0081] Figure 7 This is a structural block diagram of the terminal device according to Embodiment 3 of the present invention. Detailed Implementation

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described are merely used to explain this application and are not intended to limit this application.

[0083] Example 1:

[0084] like Figure 1 , 2 As shown, this embodiment provides a method for automatically acquiring vehicle point clouds at specific locations and angles, including the following steps:

[0085] S101. Install the radar and collect initial point cloud data.

[0086] Further, step S101 includes:

[0087] (1) Install and fix the radar at the exit of the highway toll station, and require that the radar illumination range can cover the entire toll station detection channel.

[0088] The radar is fixed at a height of h meters above the ground and collects point cloud data of approaching vehicles at a certain angle.

[0089] In this embodiment, h is taken as approximately 4 meters.

[0090] (2) Use radar to collect a frame of point cloud data when there are no cars. og .

[0091] S102. Preprocess the initial point cloud data to obtain a denoised point cloud.

[0092] This embodiment uses the DBSCAN density clustering algorithm to preprocess the initial point cloud data.

[0093] Further, step S102 includes:

[0094] (1) Set the neighborhood radius ε = 0.5 (in meters, and the length units mentioned below are all in meters), and the minimum number of points minPTS = 10.

[0095] (2) Traverse all points and calculate the number of points PTS in the ε neighborhood of each point. If PTS ≥ minPTS, then the point is the core point; otherwise, it is temporarily marked as a noise point.

[0096] (3) Treat all core points and all points in their ε-neighborhood as noise-reducing point clouds PC. dn Output the remaining points as noise points and remove them.

[0097] S103. Perform RANSAC fitting on the denoised point cloud to obtain the ground plane and the corresponding point cloud set.

[0098] Further, step S103 includes:

[0099] (1) From point cloud data PC dn Three non-collinear points are randomly selected to uniquely determine a plane M, whose equation is Ax + By + Cz + D = 0; where x, y, and z represent the x, y, and z coordinates respectively, and A, B, C, and D are the parameters of the plane equation.

[0100] (2) Traverse the point cloud PC dn For any point, calculate its distance to plane M:

[0101]

[0102] (3) Statistically, d is satisfied m ≤t m The number of points n1; t m The value is 0.1, which is the planar distance threshold in this embodiment.

[0103] (4) Repeat steps (1) to (3) above for 2000 iterations. Output the plane with the largest n1 as the surface plane, and call the set of all points whose distance to this plane satisfies d1≤t1 as P. m ,like Figure 3 As shown.

[0104] S104. Perform 3D calibration of the noise-reduced point cloud based on the ground plane.

[0105] The point cloud is calibrated for the first time in three dimensions. The surface plane is rotated to be parallel to the horizontal plane of the coordinate system and translated to z = -4, so that the surface plane equation Ax + By + Cz + D = 0 obtained in step S103 is converted into the target plane equation.

[0106] Further, step S104 includes:

[0107] (1) The surface plane is transformed into a plane equation of the form z = a0x + a1y + a2, as follows:

[0108]

[0109] (2) Calculate the three-dimensional transformation matrix T based on the plane equation. m .

[0110] Calculate the three-dimensional transformation matrix T m T m It can be decomposed into rotation matrices and translation matrices:

[0111]

[0112] The rotation matrix R and the translation matrix T are shown below:

[0113]

[0114] The parameters are calculated as follows: and

[0115] (3) Using the three-dimensional transformation matrix T m Denoising point cloud PC dn Perform a three-dimensional transformation.

[0116] Using the three-dimensional transformation matrix T m Rotation and translation noise reduction point cloud PC dn This makes the ground position approximately level with the target ground at z = -4, resulting in the point cloud PC. rotated .

[0117] S105. Based on the 3D calibrated point cloud and point cloud set, obtain the point cloud PC. masked For point cloud PCs maskedRANSAC fitting was performed to obtain a straight line in the direction of the road channel.

[0118] Furthermore, step S105 specifically includes:

[0119] (1) From point cloud PC rotate Remove Point cloud of range, then remove P m The corresponding point cloud is used to obtain the point cloud PC. masked .

[0120] (2) From point cloud data PC masked Two points with a z-coordinate difference of less than 0.1 are randomly selected to determine a straight line L with the equation (x-x0) / a=(y-y0) / b=(z-z0) / c, where x0, y0, z0, a, b, and c are all parameters of the straight line equation.

[0121] (3) Traverse the point cloud PC masked For any point, calculate its distance to line L:

[0122]

[0123] (4) Statistically, d is satisfied l ≤t l The number of points n2; t l The threshold value for the straight-line distance is 0.1 in this embodiment.

[0124] (5) Repeat steps (1) to (4) above for 2000 iterations, and output the straight line with the largest n2 as the straight line of the road channel direction. The result is as follows. Figure 4 As shown.

[0125] S106. Based on the straight line of the road channel direction, calculate the rotation matrix; use the rotation matrix to process the point cloud PC. masked Three-dimensional calibration is performed to obtain the corrected point cloud.

[0126] A second 3D calibration is performed on the point cloud to change the lane direction to the x-axis direction, so that the vehicle's driving direction is the x-axis direction.

[0127] Furthermore, step S106 specifically includes:

[0128] (1) Calculate the projection of the straight line of the road channel onto the xoy plane: y = a3x + a4.

[0129] (2) In order to transform the fitted line y=a3x+a4 into y=0, the rotation angle is calculated as θ=arctan(a3).

[0130] (3) Use the rotation matrix R2 corresponding to the rotation angle to rotate the point cloud PC maskedPerform rotational correction to obtain the corrected point cloud PC. rotated2 The matrix is ​​as follows:

[0131]

[0132] S107. Based on the corrected point cloud, obtain the lane position and save it as a mask.

[0133] This step obtains the y-coordinates of both sides of the calibrated lane and then saves the lane position as a mask.

[0134] Further, step S107 includes:

[0135] (1) Point cloud PC rotated2 Histogram statistics are performed along the y-axis. The total number of segments bin = 100. The coordinate range corresponding to each segment region is (E... i E i+1 ), count the number of point clouds that belong to this coordinate range.

[0136] (2) Find the index i of the maximum number of points in the histogram statistics. max .

[0137] (3) If i max If ≤bin / 2, then the y-coordinate of the left side of the lane is E. imax+1 -w, the y-coordinate on the right is E imax+1 ; if i max If the value is greater than bin / 2, then the y-coordinate of the left side of the lane is E. imax The y-coordinate on the right is E imax +w; where w is the width of the highway toll station exit;

[0138] In this embodiment, w is set to 4.5 meters.

[0139] (3) Save the lane position as a mask y∈(y0,y1).

[0140] S108. Obtain vehicle-free lane parameters based on mask.

[0141] Further, step S108 includes:

[0142] (1) Set the overall transformation matrix T total :

[0143]

[0144] (2) Set the overall mask M based on the mask. total :

[0145] z∈(-4,00∩x∈(0,ran x )∩y∈(y0,y1); where, ranx This is the effective radar scanning range, which is set to 80 in this embodiment.

[0146] (3) Obtain vehicle-free lane parameters based on the overall transformation matrix and the overall mask.

[0147] Specifically, step (3) includes:

[0148] (3-1) Collect point cloud data continuously for 10 frames when there are no vehicles. Preprocess the point cloud using the same method and step S102, utilizing the overall transformation matrix T. total Perform a 3D transformation on the preprocessed point cloud data to remove the overall mask M. total Point clouds outside the defined area.

[0149] (3-2) Along the vehicle's direction of travel, i.e. along x∈(0,80), divide the area into 3D grids every 10m to generate a total of 8 grid regions.

[0150] (3-3) Calculate the average z-coordinate of the point cloud in each grid within 10 frames. k0 The average x-coordinate x k0 and average maximum and minimum reflectivity R kmin R kmax , which serves as a parameter for the carless lane, where k represents the grid number.

[0151] S109. Based on the parameters of the vehicleless lane, collect vehicle point cloud data in real time.

[0152] Further, step S109 includes:

[0153] (1) Acquire a frame of point cloud data; preprocess the point cloud using the same method as step S102; utilize the overall transformation matrix T total Perform a 3D transformation on the preprocessed point cloud data to remove the overall mask M. total Point clouds outside the defined area.

[0154] (2) Segment the point cloud using the same method as in step S108 (3-2), and calculate the average z-coordinate height z of the point cloud within each grid. k Average x-coordinate depth x k and average reflectivity R k Calculate the Boolean values ​​for the data acquisition conditions:

[0155]

[0156] (3) If Collection is 0, meaning the collection condition is not met, then no changes or processing are performed and the process returns to step (1) to continue execution; if Collection is 1, meaning the collection condition is met, let x k0 =x kSave the current point cloud frame and return to step (1) to continue execution.

[0157] The partial collection results of vehicle point cloud data are as follows: Figure 5 As shown.

[0158] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0159] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0160] Example 2:

[0161] like Figure 6 As shown, this embodiment provides an automatic vehicle point cloud acquisition system for specific positions and angles. The system includes an initial acquisition module 601, a first fitting module 602, a first calibration module 603, a second fitting module 604, a second calibration module 605, a calculation module 606, and a real-time acquisition module 607, wherein:

[0162] The initial acquisition module 601 is used to acquire a frame of point cloud data when there are no vehicles as the initial point cloud data.

[0163] The first fitting module 602 is used to perform RANSAC fitting on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set.

[0164] The first calibration module 603 is used to calculate the three-dimensional transformation matrix based on the ground plane; and to perform three-dimensional calibration on the preprocessed initial point cloud data using the three-dimensional transformation matrix.

[0165] The second fitting module 604 is used to perform range filtering on the 3D calibrated point cloud based on the point cloud set to obtain the point cloud PC. masked For point cloud PCs masked Perform RANSAC fitting to obtain a straight line in the direction of the road channel;

[0166] The second calibration module 605 is used to calculate the rotation matrix based on the straight line of the road channel direction; and to use the rotation matrix to calibrate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud;

[0167] The calculation module 606 is used to perform histogram statistics on the corrected point cloud and calculate the lane position; based on the lane position, the overall mask is obtained; and based on the three-dimensional transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained.

[0168] The real-time acquisition module 607 is used to acquire point clouds in real time; it performs three-dimensional calibration and range filtering on the preprocessed point clouds based on the overall transformation matrix and the overall mask; it performs 3D mesh segmentation on the range-filtered point clouds, and selects the point cloud acquisition timing based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh, so as to realize the automatic acquisition of vehicle point cloud data.

[0169] The specific implementation of each module in this embodiment can be found in Embodiment 1 above, and will not be repeated here. It should be noted that the system provided in this embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above.

[0170] Example 3:

[0171] This embodiment provides a terminal device, which can be a computer, such as... Figure 7 As shown, the processor 702, memory, input device 703, display 704, and network interface 705 are connected via system bus 701. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium 706 and internal memory 707. The non-volatile storage medium 706 stores the operating system, computer programs, and database. The internal memory 707 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the processor 702 executes the computer programs stored in the memory, it implements the automatic vehicle point cloud acquisition method for specific positions and angles described in Embodiment 1 above, as follows:

[0172] Collect a frame of point cloud data when there are no cars as the initial point cloud data;

[0173] RANSAC fitting was performed on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set.

[0174] Based on the ground plane, calculate the three-dimensional transformation matrix; use the three-dimensional transformation matrix to perform three-dimensional calibration on the preprocessed initial point cloud data;

[0175] Based on the point cloud set, range filtering is performed on the 3D calibrated point cloud to obtain the point cloud PC. masked For point cloud PCs masked Perform RANSAC fitting to obtain a straight line in the direction of the road channel;

[0176] Calculate the rotation matrix based on the straight line of the road channel; use the rotation matrix to manipulate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud;

[0177] Histogram statistics are performed on the corrected point cloud to calculate lane positions; based on the lane positions, the overall mask is obtained.

[0178] Based on the three-dimensional transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained;

[0179] Real-time acquisition of point clouds; 3D calibration and range filtering of the preprocessed point clouds based on the overall transformation matrix and overall mask; 3D mesh segmentation of the range-filtered point clouds; selection of point cloud acquisition timing based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh to achieve automatic acquisition of vehicle point cloud data.

[0180] Example 4:

[0181] This embodiment provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the automatic vehicle point cloud acquisition method for specific positions and angles described in Embodiment 1 above, as follows:

[0182] Collect a frame of point cloud data when there are no cars as the initial point cloud data;

[0183] RANSAC fitting was performed on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set.

[0184] Based on the ground plane, calculate the three-dimensional transformation matrix; use the three-dimensional transformation matrix to perform three-dimensional calibration on the preprocessed initial point cloud data;

[0185] Based on the point cloud set, range filtering is performed on the 3D calibrated point cloud to obtain the point cloud PC. masked For point cloud PCs masked Perform RANSAC fitting to obtain a straight line in the direction of the road channel;

[0186] Calculate the rotation matrix based on the straight line of the road channel; use the rotation matrix to manipulate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud;

[0187] Histogram statistics are performed on the corrected point cloud to calculate lane positions; based on the lane positions, the overall mask is obtained.

[0188] Based on the three-dimensional transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained;

[0189] The system acquires point cloud data in real time; performs 3D calibration and range filtering on the preprocessed point cloud based on the overall transformation matrix and overall mask; performs 3D mesh segmentation on the range-filtered point cloud; and selects the timing of point cloud acquisition based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh to achieve automatic acquisition of vehicle point cloud data.

[0190] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0191] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for automatically acquiring vehicle point clouds at specific locations and angles, characterized in that, The method includes: Collect a frame of point cloud data when there are no cars as the initial point cloud data; RANSAC fitting was performed on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set. Based on the ground plane, calculate the three-dimensional transformation matrix; use the three-dimensional transformation matrix to perform three-dimensional calibration on the preprocessed initial point cloud data; Based on the point cloud set, range filtering is performed on the 3D calibrated point cloud to obtain the point cloud PC. masked For point cloud PCs masked Perform RANSAC fitting to obtain a straight line in the direction of the road channel; Calculate the rotation matrix based on the straight line of the road channel; use the rotation matrix to manipulate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud; Histogram statistics are performed on the corrected point cloud to calculate lane positions; based on the lane positions, the overall mask is obtained. Based on the three-dimensional transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained; Real-time acquisition of point clouds; 3D calibration and range filtering of the preprocessed point clouds based on the overall transformation matrix and overall mask; 3D mesh segmentation of the range-filtered point clouds; selection of point cloud acquisition timing based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh to achieve automatic acquisition of vehicle point cloud data.

2. The automatic vehicle point cloud acquisition method according to claim 1, characterized in that, Before real-time point cloud acquisition, the following is also included: Point cloud data of n consecutive frames without vehicles were acquired and preprocessed. Then, based on the overall transformation matrix and overall mask, 3D calibration and range filtering were performed on the preprocessed point cloud data. Next, the range-filtered point cloud was divided into 3D meshes along the vehicle's driving direction, and the average z-coordinate z of the point cloud within each 3D mesh of the n frames was calculated. k0 Average x-coordinate k0 and average maximum and minimum reflectivity R kmin R kmax Where k represents the grid number, and n is greater than or equal to 2 and less than or equal to ran. x A positive integer of 3, ran x This represents the effective range of the radar scan.

3. The automatic vehicle point cloud acquisition method according to claim 2, characterized in that, The real-time acquisition of point clouds; the 3D calibration and range filtering of the preprocessed point clouds based on the overall transformation matrix and overall mask; the 3D mesh segmentation of the range-filtered point clouds; and the selection of point cloud acquisition timing based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh, including: A frame of point cloud data is acquired and preprocessed; then, based on the overall transformation matrix and overall mask, 3D calibration and range filtering are performed on the preprocessed point cloud data; finally, the range-filtered point cloud is divided into 3D meshes along the vehicle's driving direction, and the average z-coordinate height z of the point cloud within each 3D mesh is calculated. k Average x-coordinate depth x k and average reflectivity R k ; Calculate the Boolean value of the acquisition conditions If Collection is 0, then return to collect one frame of point cloud data, perform preprocessing, and continue with subsequent operations; if Collection is 1, let x... k0 =x k Save the current point cloud frame, return to collect a frame of point cloud data, perform preprocessing, and continue with subsequent operations.

4. The automatic vehicle point cloud acquisition method according to any one of claims 1 to 3, characterized in that, The step of performing histogram statistics on the corrected point cloud and calculating the lane position includes: Histogram statistics were performed on the corrected point cloud along the y-axis: a total number of segments (bin) was set, and the coordinate range corresponding to each segmented region was (E... i E i+1 ), count the number of point clouds that fall within this coordinate range; Find the index i of the maximum number of point clouds in the histogram statistics. max ; If i max If ≤bin / 2, then the y-coordinate of the left side of the lane is E. imax+1 -w, the y-coordinate of the right side of the lane is E imax+1 ; if i max If >bin / 2, then the y-coordinate of the left side of the lane is E. imax The y-coordinate of the right side of the lane is E. imax +w; where w is the standard toll station exit width.

5. The automatic vehicle point cloud acquisition method according to any one of claims 1 to 3, characterized in that, The equation of the surface plane is A0x + B0y + C0z + D0 = 0; where x, y, and z represent the x, y, and z coordinates respectively, and A0, B0, C0, and D0 are plane equation parameters. The calculation of the three-dimensional transformation matrix based on the ground plane includes: The surface plane is transformed into a plane equation of the form z = a0x + a1y + a2, as follows: Based on the plane equation, we obtain the rotation matrix R and the translation matrix T: in, h represents the height above the ground when the radar acquires the initial point cloud data; Based on the rotation matrix R and the translation matrix T, the three-dimensional transformation matrix is ​​obtained.

6. The automatic vehicle point cloud acquisition method according to any one of claims 1 to 3, characterized in that, The calculation of the rotation matrix based on the straight line of the road channel direction includes: Let the projection of the straight line of the road channel onto the xoy plane be y = a3x + a4; To transform the line y = a3x + a4 into y = 0, calculate the rotation angle θ = arctan(a3); Based on the rotation angle, the corresponding rotation matrix is ​​obtained.

7. The automatic vehicle point cloud acquisition method according to any one of claims 1 to 3, characterized in that, The step of performing RANSAC fitting on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set includes: From the preprocessed initial point cloud data PC dn Three non-collinear points are randomly selected to uniquely determine plane M; the equation of plane M is Ax + By + Cz + D = 0, where x, y, and z represent the x, y, and z coordinates respectively, and A, B, C, and D are parameters of the plane equation. Traversing point clouds PC dn Given all points, calculate the distance from any point to plane M: Statistical satisfaction d m ≤t m The number of points n1; t m The set planar distance threshold; After iterating a specified number of times, the plane corresponding to the largest n1 is taken as the surface plane, and the plane with a distance ≤t from the surface plane is considered as the surface plane. m The set of all points is called the point cloud set.

8. The automatic vehicle point cloud acquisition method according to any one of claims 1 to 3, characterized in that, The point cloud PC masked Perform RANSAC fitting to obtain the straight line in the road channel direction, including: From point cloud data PC masked Two points with a z-coordinate difference less than a specified value are randomly selected to determine the line L; the equation of the line L is (x-x0) / a=(y-y0) / b=(z-z0) / c, where x0, y0, z0, a, b, and c are all parameters of the line equation. Traversing point clouds PC masked Given all points, calculate the distance from any point to line L: Statistical satisfaction d l ≤t l The number of points n2; t l The specified straight-line distance threshold; Iterate a specified number of times, and take the line corresponding to the maximum n2 as the straight line of the road channel direction.

9. The automatic vehicle point cloud acquisition method according to any one of claims 1 to 3, characterized in that, The preprocessing includes: Set the neighborhood radius ε and the minimum number of points minPTS; Traverse all points in the point cloud and calculate the number of points PTS in the ε-neighborhood of each point; if PTS ≥ minPTS, then the point is a core point; otherwise, temporarily mark it as a noise point. All core points and all points within their ε-neighborhood are output as denoised point clouds, and the remaining points are removed as noise points.

10. An automatic point cloud acquisition system for vehicles at specific locations and angles, characterized in that, The system includes: The initial acquisition module is used to acquire a frame of point cloud data when there are no vehicles as the initial point cloud data. The first fitting module is used to perform RANSAC fitting on the preprocessed initial point cloud data to obtain the land surface plane and the corresponding point cloud set. The first calibration module is used to calculate the three-dimensional transformation matrix based on the ground plane; and to perform three-dimensional calibration on the preprocessed initial point cloud data using the three-dimensional transformation matrix. The second fitting module is used to perform range filtering on the 3D calibrated point cloud based on the point cloud set to obtain the point cloud PC. masked For point cloud PCs masked Perform RANSAC fitting to obtain a straight line in the direction of the road channel; The second calibration module is used to calculate the rotation matrix based on the straight line of the road channel direction; and to use the rotation matrix to calibrate the point cloud PC. masked Perform 3D calibration to obtain the corrected point cloud; The calculation module is used to perform histogram statistics on the corrected point cloud and calculate the lane position; based on the lane position, the overall mask is obtained; and based on the 3D transformation matrix and rotation matrix, the overall transformation matrix is ​​obtained. The real-time acquisition module is used to acquire point clouds in real time; based on the overall transformation matrix and overall mask, the preprocessed point cloud is then subjected to 3D calibration and range filtering; the range-filtered point cloud is segmented into a 3D mesh, and the timing of point cloud acquisition is selected based on the changes in the centroid spatial coordinates and average reflectivity of the 3D mesh, so as to realize the automatic acquisition of vehicle point cloud data.

Images