Multi-view laser radar calibration and point cloud fusion method, device and storage medium

Abstract

The application discloses a multi-view laser radar automatic calibration and point cloud fusion method, equipment and a storage medium; the method comprises the following steps: transforming an auxiliary view field coordinate system into a main view field coordinate system through a first transformation matrix; converting a ground point in a radar coordinate system into a point cloud in a field coordinate system by the ground point, a second rotation matrix and a second translation vector; and superimposing and fusing the point clouds after multi-view fine registration to obtain fusion point clouds in the main view field coordinate system. By setting multiple laser radars and unifying the coordinates from the main view, the accurate alignment of different view data is ensured, and the overall measurement accuracy is improved. The second rotation matrix and the second translation vector determined in combination with the ground point are used for coordinate conversion, and the robustness of the algorithm is enhanced. The point clouds of all views converted into the main view field coordinate system are superimposed and fused, the complementarity of multi-view data is fully utilized, and the completeness and detail richness of the point cloud are improved.

Description

Technical Field

[0001] This application relates to the field of target calibration technology, and in particular to a method, device and storage medium for automatic calibration and point cloud fusion of multi-view lidar. Background Technology

[0002] Data analysis plays a significant tactical and technical role in multi-player sports, with athletes' movement trajectories being the most fundamental and crucial component. Traditional camera-based or wearable device-based solutions struggle to simultaneously meet the requirements for high-precision and seamless data acquisition.

[0003] Using other types of sensing devices is one way to improve accuracy, achieve automation, and maintain seamless tracking. Advances in technologies such as microelectromechanical systems (MEMS) or rotating prisms have made lidar a promising sensor for many applications, including autonomous driving, mapping, and target tracking.

[0004] Currently, the vast majority of LiDAR calibration work originates from autonomous driving or environmental perception applications (mapping, navigation, robotics, etc.). In these scenarios, LiDAR scans from the center outwards. Because LiDARs are placed close to each other, different LiDARs can automatically detect the same features from areas of common coverage. For example, manual landmarks or environmental appearance cues are used to determine the spatial offset of multiple LiDARs, thereby achieving automated matching.

[0005] When detecting targets (athletes or referees, etc.) on a sports field, due to the limitations of the effective detection field of view and effective detection range of lidar, it is known that as the distance between the target and the lidar increases, the number of points falling on a unit volume of target decreases exponentially. Therefore, a single lidar is insufficient for scenarios requiring long-distance, large-area target perception (such as formal football or rugby matches). Thus, it is necessary to use fused perception from multiple lidars to comprehensively acquire information about the field.

[0006] Multiple lidars capture the same area from different locations. Although the point clouds from lidars with different perspectives also cover the same area, they cannot be directly matched for matching because they are perceived from different perspectives and the lidars are too far apart. Therefore, automatic calibration is not possible. Summary of the Invention

[0007] The main technical problem addressed in this application is to provide a method for automatic calibration and point cloud fusion of multi-view LiDAR, which solves the problem that automatic calibration is not possible when sensing from different perspectives.

[0008] To address the aforementioned technical problems, this application provides a method, device, and storage medium for automatic calibration and point cloud fusion of multi-view lidar, comprising the following steps:

[0009] Step 1): Around the sports field, multiple lidars are installed. One lidar is randomly selected as the main view lidar and the other lidars are used as auxiliary view lidars. The main view field coordinate system and the auxiliary view field coordinate system are established. The auxiliary view field coordinate system is transformed into the main view field coordinate system through the first transformation matrix.

[0010] Step 2): Extract the point cloud in the lidar coordinate system using a point cloud extraction algorithm to obtain ground points in the lidar coordinate system. Determine the second rotation matrix and the second translation vector from the ground points. Convert the ground points in the lidar coordinate system into a point cloud in the site coordinate system using the ground points, the second rotation matrix, and the second translation vector.

[0011] Step 3): Based on the point clouds corresponding to the LiDARs from different viewpoints in the site coordinate system, and combining the first rotation matrix and the first translation vector, determine whether the auxiliary viewpoint site coordinate system has been converted to the main viewpoint site coordinate system. If so, the point clouds of all viewpoints converted to the main viewpoint site coordinate system are superimposed and fused. The target in the point clouds of all viewpoints is detected by the neural network, and it is determined whether the point clouds of the auxiliary viewpoint LiDARs have completed fine registration using the target position. If so, the point clouds after fine registration of multiple viewpoints are superimposed and fused to obtain the fused point cloud of all viewpoints located in the main viewpoint site coordinate system.

[0012] This application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-view lidar automatic calibration and point cloud fusion method.

[0013] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the multi-view lidar automatic calibration and point cloud fusion method.

[0014] The beneficial effects of this application are as follows: By setting up multiple lidars and unifying coordinates from the main viewpoint, the first transformation matrix is ​​used to accurately transform the auxiliary viewpoint field coordinate system to the main viewpoint field coordinate system, ensuring precise alignment of data from different viewpoints and improving overall measurement accuracy. Combining the second rotation matrix and second translation vector determined by the ground points for coordinate transformation enhances the robustness of the algorithm. The point clouds from all viewpoints transformed into the main viewpoint field coordinate system are superimposed and fused, fully utilizing the complementarity of multi-viewpoint data and improving the integrity and detail richness of the point cloud. Fine registration further optimizes the point cloud data, ensuring that the fused point cloud is more accurate and consistent, providing a high-quality data foundation for subsequent motion analysis and processing. Attached Figure Description

[0015] Figure 1 This is a layout diagram of a multi-view LiDAR system in a multi-person motion scene according to an embodiment of this application;

[0016] Figure 2 This is a network connection diagram of a multi-view lidar according to an embodiment of this application;

[0017] Figure 3 It is a reference coordinate system from the point cloud perspective according to an embodiment of this application;

[0018] Figure 4 This is a schematic diagram illustrating the transformation process from the secondary viewpoint site coordinate system to the primary viewpoint site coordinate system according to an embodiment of this application.

[0019] Figure 5 This is a flowchart of a single lidar calibration process according to an embodiment of this application;

[0020] Figure 6 This is a flowchart of a multi-view lidar fusion process according to an embodiment of this application. Detailed Implementation

[0021] To facilitate understanding of this application, a more detailed description is provided below with reference to the accompanying drawings and specific embodiments. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0022] It should be noted that, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0023] Figures 1-6 An embodiment of the multi-view lidar automatic calibration and point cloud fusion method of this application is shown, including:

[0024] Step 1): Around the sports field, multiple lidars are installed. One lidar is randomly selected as the main view lidar and the other lidars are used as auxiliary view lidars. The main view field coordinate system and the auxiliary view field coordinate system are established. The auxiliary view field coordinate system is transformed into the main view field coordinate system through the first transformation matrix.

[0025] Step 2) Extract the point cloud in the lidar coordinate system using a point cloud extraction algorithm to obtain ground points in the lidar coordinate system. Determine the second rotation matrix and the second translation vector from the ground points. Use the ground points, the second rotation matrix, and the second translation vector to convert the ground points in the lidar coordinate system into a point cloud in the site coordinate system.

[0026] Step 3): Based on the point clouds corresponding to the LiDARs from different viewpoints in the site coordinate system, and combining the first rotation matrix and the first translation vector, determine whether the auxiliary viewpoint site coordinate system has been converted to the primary viewpoint site coordinate system. If so, overlay and fuse the point clouds of all viewpoints converted to the primary viewpoint site coordinate system. Detect targets in the point clouds of all viewpoints using a neural network, and determine whether the point clouds of the auxiliary viewpoint LiDARs have completed fine registration using the target positions. If so, overlay and fuse the finely registered point clouds of multiple viewpoints to obtain a fused point cloud of all viewpoints located in the primary viewpoint site coordinate system.

[0027] In this application, multiple lidar sensors are deployed, and coordinates are unified from the primary viewpoint. A first transformation matrix is ​​used to accurately transform the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system, ensuring precise alignment of data from different viewpoints and improving overall measurement accuracy. Combining the second rotation matrix and second translation vector determined by ground points for coordinate transformation enhances the algorithm's robustness. The point clouds from all viewpoints transformed into the primary viewpoint field coordinate system are then superimposed and fused, fully utilizing the complementarity of multi-viewpoint data and improving the integrity and detail richness of the point cloud. Fine registration further optimizes the point cloud data, ensuring the fused point cloud is more accurate and consistent, providing a high-quality data foundation for subsequent motion analysis and processing.

[0028] In step 1), such as Figure 1 and Figure 2Multiple lidar sensors 1 are installed around the sports field 2. One lidar sensor 1 is randomly selected as the primary view lidar, and the other lidar sensors are used as secondary view lidars. Each lidar sensor 1 has a corresponding reference point. The primary view field coordinate system is established with the reference point of the primary view lidar as the origin. The secondary view field coordinate system is established with the reference point of the secondary view lidar as the origin. The secondary view field coordinate system is transformed into the primary view field coordinate system through a first transformation matrix.

[0029] Sports field 2 can be a football field, basketball court, badminton court, etc. A lidar 1 can be installed at each of the four corner flag points (four corners) of the sports field. Lidar 1 can also be installed along the sidelines of the sports field. One lidar 1 can be randomly selected as the primary view lidar, and the other lidars can be used as secondary view lidars.

[0030] The nearest corner flag point to the lidar can be used as a reference point, and this reference point can be used as the origin to establish the site coordinate system. The site coordinate system established by the main-view lidar is the main-view site coordinate system. The site coordinate system established by the secondary-view lidar is the secondary-view site coordinate system. For example... Figure 3 As shown, L LiDAR The location of the lidar is shown on the X-axis. LiDAR Y LiDAR Z LiDAR The reference point for the lidar is L. Corner Reference point L Corner The corresponding coordinate axes are X and X. Corner Y Corner Z Corner .

[0031] Because the primary and secondary view lidars are located at different positions, their respective site coordinate systems are also different. This results in a lack of coordinate system unification for the lidars, preventing the provision of a reliable spatial reference in subsequent analyses. Therefore, it is necessary to unify the primary and secondary view site coordinate systems.

[0032] Using the main viewpoint field coordinate system as the reference, the auxiliary viewpoint field coordinate system is converted into the main viewpoint field coordinate system to unify the coordinate system of the lidar.

[0033] Establish the first transformation matrix T ini T ini Represented as:

[0034] T ini ={R ini , t ini s ini}

[0035] Among them, Rini Let t be the first rotation matrix from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system. ini This refers to the first translation vector from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system. ini It refers to the vertical movement vector from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system.

[0036] In some embodiments, the lidar units can be positioned at the same height, thereby eliminating the influence of different vertical positions of the lidar units and reducing the computational load. In this case, T ini It can be represented as:

[0037] T ini ={R ini , t ini}

[0038] The first rotation matrix is ​​formed because the primary viewpoint field coordinate system and the secondary viewpoint field coordinate system are not in the same direction, so rotation is required. The rotation angle is 90° or a multiple of it (180°, 270°) around the z-axis.

[0039] The formation of the first translation vector is due to the fact that the main view field coordinate system and the secondary view field coordinate system are not in the same position. The first translation vector corresponds to the length of the sideline or baseline of the sports field.

[0040] The secondary viewpoint field coordinate system is transformed into the primary viewpoint field coordinate system through the first transformation matrix, thereby unifying the coordinate systems of all lidar sensors.

[0041] Specifically, such as Figure 4 As shown, As the main perspective site coordinate system As a secondary perspective site coordinate system When transforming to the main viewpoint field coordinate system, the first transformation matrix is: It is 90°. Let be the vector of the baseline of the sports field. When transforming to the main viewpoint field coordinate system, the first transformation matrix is: It is 180°. is the vector of the diagonal of the sports field. When transforming to the main viewpoint field coordinate system, the first transformation matrix is: It is 270°. Let be the vector of the sports field boundary line.

[0042] The above methods can unify the coordinate system of the lidar, allowing for the overlay and analysis of multi-view data, thereby improving the accuracy of tasks such as target detection and 3D reconstruction.

[0043] In step 2), the point cloud in the lidar coordinate system is extracted using a point cloud extraction algorithm to obtain ground points in the lidar coordinate system. The ground points are used to determine the second rotation matrix and the second translation vector. The ground points, the second rotation matrix, and the second translation vector are used to convert the ground points in the lidar coordinate system into a point cloud in the site coordinate system.

[0044] Specific steps are as follows Figure 5 As shown:

[0045] Let P be the point cloud in the lidar coordinate system. L ;

[0046] Point cloud extraction algorithm is used to extract point cloud P in lidar coordinate system L Extract the ground point P that belongs to the ground. L-plane Point cloud extraction algorithms include RANSAC algorithm or Hough transform, etc.

[0047] Using ground point P L-plane Obtain the ground normal vector of the lidar coordinate system a represents the first element of the normal vector, and b represents the second element of the normal vector.

[0048] The ground normal vector is rotated and used as the Z-axis of the field coordinate system. Then, the direction vector of one of the straight lines is used as the y-axis direction. The x-axis direction is obtained by the cross product of the z-axis and y-axis directions.

[0049] From the ground normal vector Determine the first intermediate rotation matrix R1 and represent the first intermediate rotation matrix as follows:

[0050] The point cloud in the lidar coordinate system is rotated using the first intermediate rotation matrix R1 to obtain the first ground point P. 1-plane ;

[0051] For the first ground point P 1-plane Take the average value to obtain the average value zMean1;

[0052] By uniformly subtracting the average value zMean1 in the z-direction, the ground point is translated to the xoy plane, resulting in a new first ground point P. 1-plane ;

[0053] Range filtering is performed on the point cloud in the lidar coordinate system to obtain the second ground point P closest to the corner flag point. 2-plane ;

[0054] Intensity filtering is performed on the point cloud in the lidar coordinate system to obtain the two sides of the corner flag point – the goal line and the sideline point P. 3-plane ;

[0055] Two point cloud extraction algorithms were performed to obtain the goal line (line1), sideline points (line2), and vectors in the sports field.

[0056] The goal line (line1), the sideline point (line2), and the vector... Determine the corner flag point O in the site coordinate system;

[0057] From vector Determine the second intermediate rotation matrix R2, which is expressed as follows: If but like but The second intermediate rotation matrix is ​​a 3x3 matrix, where each row is a 1x3 vector. These represent the vectors in the first, second, and third rows of this rotation matrix, respectively.

[0058] From the corner flag point O in the field coordinate system, obtain the first ground point P. 1-plane The second intermediate translation vector t2 is expressed as: t2 = -1 × [O] T ;T indicates transpose.

[0059] The second rotation matrix R is obtained from the first intermediate rotation matrix R1 and the second intermediate rotation matrix R2. {L→F} , represented as R {L→F} =R2×R1;

[0060] The second translation vector t is obtained from the second intermediate rotation matrix R2, the first intermediate translation vector t1, and the second intermediate translation vector t2. {L→R} , represented as: t {L→R} =R2×(t1+t2); The first intermediate translation vector t1 is obtained by physically measuring the lengths of the sidelines and spherical lines of the football field.

[0061] Finally, by the second rotation matrix R {L→F} Ground point P L-plane Second translation vector t {L→R} Ground point P in radar coordinate system L-plane Convert to point cloud P in site coordinate system F , represented as: P F =R {L→R} ×P L-plane +t {L→R} .

[0062] In step 3), such as Figure 6 As shown, based on the point clouds corresponding to the LiDARs from different viewpoints in the site coordinate system, and combining the first rotation matrix and the first translation vector, it is determined whether the auxiliary viewpoint site coordinate system has been converted to the primary viewpoint site coordinate system. If the auxiliary viewpoint site coordinate system has been converted to the primary viewpoint site coordinate system, the point clouds of all viewpoints converted to the primary viewpoint site coordinate system are superimposed and fused. A neural network is then used to detect targets in all viewpoint point clouds to determine whether the point clouds of the auxiliary viewpoint LiDARs have completed fine registration using the target position. If the point clouds of the LiDARs have completed fine registration using the target position, the point clouds of multiple viewpoints after fine registration are superimposed and fused to obtain a fused point cloud of all viewpoints located in the primary viewpoint site coordinate system.

[0063] The point clouds of the lidar units in their respective site coordinate systems can be obtained using the methods described above; each lidar unit corresponds to one point cloud. Point clouds from different viewpoints in their respective site coordinate systems are shown. The point cloud is represented as an N x 4 matrix, where N represents the number of points in the point cloud, and 4 represents that each point has 4 features. These features are all real numbers. m represents the number of lidar units. The first rotation matrix from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system. SO denotes a special orthogonal group, consisting of all orthogonal square matrices of the same order with a determinant of 1, and is the set of all rotation matrices. The first translation vector from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system.

[0064] When determining whether the secondary viewpoint site coordinate system has been converted to the primary viewpoint site coordinate system; if the secondary viewpoint site coordinate system has not been converted to the primary viewpoint site coordinate system, then determine whether the point cloud corresponds to the primary viewpoint site coordinate system. If it is the primary viewpoint site coordinate system, then perform coarse registration to convert other secondary viewpoint site coordinate systems to the primary viewpoint site coordinate system. The point cloud represents the coarse transformation. This represents the translation vector during coarse registration. If it's not the main view field coordinate system, no transformation is needed.

[0065] If the auxiliary viewpoint site coordinate system is converted to the primary viewpoint site coordinate system, then the point clouds of all views in the primary viewpoint site coordinate system will be superimposed and merged, as shown below: This represents the fused point cloud after coarse registration.

[0066] Spatial filtering is performed on the point clouds from all viewpoints to obtain filtered point clouds. Filtered point cloud Represented as The points filter indicates that clutter filtering is applied to the point cloud, including clutter from the ground and outside the field.

[0067] Filtered point cloud from all viewpoints Target point detection is performed using a neural network to obtain detection results. The neural network can be a cascaded network, such as PointNet or PointNet++. Detection results. Represented as: Cascade PointNet represents a cascaded network that performs target recognition on filtered point clouds, identifying standing targets within a sports field and their center of gravity.

[0068] Determine whether the point cloud of the secondary view lidar has completed fine registration using the target position.

[0069] If the point cloud from the secondary view LiDAR has not been finely registered using the target position, then the cost matrix is ​​obtained by using the distance between the detection results from the secondary view LiDAR and the detection results from the primary view LiDAR. cdist represents the Euclidean distance between the target set from the auxiliary viewpoint and the target set from the primary viewpoint.

[0070] Matches are obtained using the Hungarian algorithm. i =hungarian(diatMatrix) i ), i∈[2,m]; Hungarian indicates that the matching relationship between the target set under the secondary view and the target set under the primary view is calculated using the Hungarian algorithm based on the distance calculated in the previous step.

[0071] The ICP algorithm is used to obtain the fine-tuning rotation matrix and translation vector for fine registration of the matching strategy. ICP indicates that, based on the matching relationship calculated in the previous step, the ICP algorithm is used to obtain the fine registration rotation matrix. and fine registration translation vector Fine registration results for the secondary viewpoint are obtained from the fine registration rotation matrix and fine registration translation vector. Then, check again whether the point cloud of the secondary view lidar has completed fine registration using the target position. This step can be repeated until the point cloud of the secondary view lidar has completed fine registration using the target position.

[0072] If the point cloud from the secondary view LiDAR is finely registered using the target position, then the point clouds from the primary view LiDAR and the secondary view LiDAR are superimposed and fused. At this point, the total rotation matrix from the secondary view field coordinate system to the primary view field coordinate system is expressed as: The total translation vector from the secondary viewpoint site coordinate system to the primary viewpoint site coordinate system is expressed as: The point cloud P, obtained by fusing multiple viewpoints in the main field coordinate system, can be obtained from the total rotation matrix and the total translation vector. combined ; Fusion point cloud P combined Represented as:

[0073] In some embodiments, such as Figure 2 As shown, the fused point cloud can be uploaded to the cloud server 4 through the edge server 3. The cloud server 4 can analyze the motion state of the target in the sports field 2 based on the fused point cloud. The smart terminal 5 can connect to the cloud server 4 through the network and access the fused point cloud or motion state therein.

[0074] In this application, multiple lidar sensors are deployed, and coordinates are unified from the primary viewpoint. A first transformation matrix is ​​used to accurately transform the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system, ensuring precise alignment of data from different viewpoints and improving overall measurement accuracy. Combining the second rotation matrix and second translation vector determined by ground points for coordinate transformation enhances the algorithm's robustness. The point clouds from all viewpoints transformed into the primary viewpoint field coordinate system are then superimposed and fused, fully utilizing the complementarity of multi-viewpoint data and improving the integrity and detail richness of the point cloud. Fine registration further optimizes the point cloud data, ensuring the fused point cloud is more accurate and consistent, providing a high-quality data foundation for subsequent motion analysis and processing.

[0075] The above are merely embodiments of this application and do not limit the scope of this patent application. Any equivalent structural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A method for automatic calibration and point cloud fusion of multi-view lidar, characterized in that, Including the following steps: Step 1) Around the sports field, multiple lidars are installed. One lidar is randomly selected as the main view lidar and the others are used as auxiliary view lidars. The main view field coordinate system and the auxiliary view field coordinate system are established. The auxiliary view field coordinate system is transformed into the main view field coordinate system through the first transformation matrix. The lidar is positioned at the same height, and the first transformation matrix... Represented as: ； in, This refers to the first rotation matrix from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system. The first translation vector from the auxiliary viewpoint field coordinate system to the main viewpoint field coordinate system; Step 2) Extract the point cloud in the lidar coordinate system using a point cloud extraction algorithm to obtain the ground points in the lidar coordinate system. Determine the second rotation matrix and the second translation vector from the ground points. Use the ground points, the second rotation matrix, and the second translation vector to convert the ground points in the lidar coordinate system into a point cloud in the site coordinate system. Step 3) Based on the point clouds of the LiDAR at different viewpoints in the site coordinate system, and combining the first rotation matrix and the first translation vector, determine whether the auxiliary viewpoint site coordinate system has been converted into the main viewpoint site coordinate system; if so, then the point clouds of all viewpoints converted into the main viewpoint site coordinate system are superimposed and fused; by detecting targets in the point clouds of all viewpoints through a neural network, determine whether the point clouds of the auxiliary viewpoint LiDAR have completed fine registration using the target position; if so, then the point clouds after fine registration of multiple viewpoints are superimposed and fused to obtain a fused point cloud of all viewpoints located in the main viewpoint site coordinate system.

2. The method for automatic calibration and point cloud fusion of multi-view lidar according to claim 1, characterized in that, Let the point cloud in the lidar coordinate system be denoted as... The point cloud extraction algorithm is used to extract the point cloud in the lidar coordinate system. Extraction is performed to obtain ground points belonging to the ground. ; Utilizing ground points Obtain the ground normal vector of the lidar coordinate system Where a represents the first element of the normal vector and b represents the second element of the normal vector; the coordinate axes of the site coordinate system are determined by the ground normal vector.

3. The method for automatic calibration and point cloud fusion of multi-view lidar according to claim 2, characterized in that, From the ground normal vector Determine the first intermediate rotation matrix, and represent the first intermediate rotation matrix as follows: ; Using the first intermediate rotation matrix Rotate the point cloud in the lidar coordinate system to obtain the first ground point. ; Two point cloud extraction algorithms were performed to obtain the goal lines in the sports field. , edge point and vectors ; From the goal line , edge point and vectors Determine the corner flag point O in the aforementioned site coordinate system; From the vector Determine the second intermediate rotation matrix The second intermediate rotation matrix Represented as If >0, , ;like ≤0, then , ; , , These represent the three row vectors of the second intermediate rotation matrix; From the corner flag point O in the site coordinate system, the first ground point is obtained. The second intermediate translation vector below The second intermediate translation vector Represented as: T indicates transpose; From the first intermediate rotation matrix Second intermediate rotation matrix Obtain the second rotation matrix The second rotation matrix Represented as ; By the second intermediate rotation matrix First intermediate translation vector Second intermediate translation vector Obtain the second translation vector The second translation vector Represented as: The first intermediate translation vector These are the measurements for the sports field; By the second rotation matrix Ground point Second translation vector The ground points in the laser radar coordinate system Convert to point cloud in site coordinate system , is represented as: .

4. The method for automatic calibration and point cloud fusion of multi-view lidar according to claim 3, characterized in that, Determine whether the auxiliary viewpoint field coordinate system has been converted to the main viewpoint field coordinate system; If the secondary viewpoint site coordinate system has not been converted to the primary viewpoint site coordinate system, then it is determined whether the coordinate system corresponding to the point cloud is the primary viewpoint site coordinate system. If it is the primary viewpoint site coordinate system, then coarse registration is performed to convert the other secondary viewpoint site coordinate systems to the primary viewpoint site coordinate system. ; The point cloud represents the coarse transformation. This represents an orthogonal value used when transforming the auxiliary viewpoint site coordinate system to the main viewpoint site coordinate system. This represents the point cloud in the site coordinate system corresponding to the i-th lidar, where i represents the index of the lidar and m represents the total number of lidars. This represents the translation vector during coarse registration; if it is not the main view field coordinate system, the main view field coordinate system does not need to be transformed. ; If the auxiliary viewpoint site coordinate system is converted to the primary viewpoint site coordinate system, then the point clouds of all views converted to the primary viewpoint site coordinate system will be superimposed and fused, as shown below: ; This represents the fused point cloud after coarse registration. This indicates an overlay operation.

5. The method for automatic calibration and point cloud fusion of multi-view lidar according to claim 4, characterized in that, Spatial filtering is performed on the point clouds from all viewpoints to obtain filtered point clouds. Filtered point cloud Represented as ; This indicates that clutter filtering is applied to the point cloud, and the filtered point cloud is applied across all viewpoints. Target point detection is performed using a neural network to obtain detection results. .

6. The method for automatic calibration and point cloud fusion of multi-view lidar according to claim 5, characterized in that, Determine whether the point cloud of the secondary view LiDAR has completed fine registration using the target position; if the point cloud of the secondary view LiDAR has not completed fine registration using the target position, then use the distance to obtain the cost matrix of the detection results of the secondary view LiDAR and the detection results of the primary view LiDAR. cdist indicates the calculation of the Euclidean distance between the target set under the auxiliary view and the target set under the main view. Matching strategy obtained from Hungarian algorithm ; Hungarian indicates that, based on the distance calculated in the previous step, the Hungarian algorithm is used to calculate the matching relationship between the target set under the secondary view and the target set under the primary view. The ICP algorithm is used to obtain the fine-tuning rotation matrix and translation vector for fine registration of the matching strategy. ICP This indicates that, based on the matching relationship calculated in the previous step, the ICP algorithm is used to obtain the fine registration rotation matrix. ; and fine registration translation vector ; Fine registration results for the secondary viewpoint are obtained from the fine registration rotation matrix and fine registration translation vector. Then, it is determined again whether the point cloud of the secondary view lidar has completed fine registration using the target position; the ICP algorithm steps are repeated until the point cloud of the secondary view lidar has completed fine registration using the target position.

7. The method for automatic calibration and point cloud fusion of multi-view lidar according to claim 6, characterized in that, If the point cloud of the secondary view LiDAR is successfully registered using the target position, then the point cloud of the primary view LiDAR and the point cloud of the secondary view LiDAR are superimposed and fused. At this point, the total rotation matrix from the secondary view site coordinate system to the primary view site coordinate system is represented as follows: The total translation vector from the secondary viewpoint field coordinate system to the primary viewpoint field coordinate system is expressed as: ; The point cloud, obtained by fusing multiple viewpoints in the main field coordinate system, can be obtained from the total rotation matrix and the total translation vector. ; Fusion point cloud Represented as: .

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-view lidar automatic calibration and point cloud fusion method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-view lidar automatic calibration and point cloud fusion method according to any one of claims 1 to 7.

Images