Calibration method and system for multiple radar extrinsic parameters of unmanned mining vehicles

Abstract

The present invention relates to the field of intelligent gateway technology, and more specifically, to a method and system for calibrating the external parameters of multiple radars of unmanned mining vehicles. The method comprises the following steps: taking a foam rectangular calibration plate as the core and designing a new calibration plate frame in conjunction with a corner reflector, and at the same time designing a miniaturized calibration field based on the calibration plate frame that can meet the task of calibrating the external parameters of multiple laser radars and millimeter-wave radars of unmanned mining vehicles; recording laser radar data of the miniaturized calibration field to obtain laser radar point cloud data in the calibration field, and realizing calibration target extraction and constraint extraction based on the calibration field point cloud data. The present invention is applicable to the calibration of multiple laser radars and millimeter waves of unmanned mining vehicles, improves efficiency, and thus solves the problems of poor accuracy of the calibration process of unmanned mining vehicles in rugged outdoor environments, insufficient global optimization and scalability, etc. in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent gateways, and in particular to a method and system for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle. Background Art

[0002] With the continuous advancement of mining technology, unmanned mining vehicles have gradually become an important component of mine automation. Unmanned mining vehicles are typically equipped with multiple sensors, such as lidar and millimeter-wave radar, to achieve high-precision perception and detection of complex mining environments. Lidar provides high-precision three-dimensional point cloud data, while millimeter-wave radar provides stable detection capabilities even in adverse weather conditions. Multi-sensor data fusion can significantly enhance the environmental perception capabilities of unmanned mining vehicles. However, multi-sensor data fusion requires precise calibration of the extrinsic parameters (i.e., relative position and attitude) between each sensor. The accuracy of this extrinsic parameter calibration directly affects the effectiveness of data fusion and, in turn, the overall performance of the unmanned mining vehicle.

[0003] The unique characteristics of mining environments pose significant challenges to traditional calibration methods. Mining operations typically involve limited space, complex terrain, and high demands for continuous production. Traditional multi-lidar and millimeter-wave radar calibration methods struggle to adapt to these conditions. Existing calibration technologies are primarily categorized as target-based and environmental feature-based. The former requires a high-quality calibration site, while the latter suffers from insufficient accuracy in mining environments due to sparse features. Furthermore, traditional calibration methods typically require complex equipment and long calibration times, making them difficult to meet the continuous and efficient demands of mining operations. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle, so as to solve the problems raised in the above background technology, such as high requirements for the calibration site, insufficient accuracy due to sparse features in the mining environment, and the need for complex equipment and long calibration time.

[0005] To achieve the above object, the present invention provides a method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle, comprising the following steps:

[0006] S1. Design a novel calibration plate framework using a foam rectangular calibration plate as the core and in conjunction with a corner reflector. Based on this calibration plate framework, design a miniaturized calibration field capable of calibrating the external parameters of multiple lidars and millimeter-wave radars for unmanned mining vehicles.

[0007] S2. Record the lidar data of the miniaturized calibration field to obtain the lidar point cloud data in the calibration field, and extract the calibration target and constraint based on the calibration field point cloud data;

[0008] S3. Establish the constraint relationship between each laser radar and the miniaturized calibration field point cloud data on the calibration target, and optimize and solve the rotation matrix With translation vector ;

[0009] S4. Recording laser radar and millimeter-wave radar data for the miniaturized calibration field to obtain laser radar point cloud data and millimeter-wave radar perception data of the calibration field, and extracting the centroid information of the laser radar and millimeter-wave radar calibration targets based on the calibration field perception data;

[0010] S5. Establish the constraint relationship between the lidar and millimeter-wave radar on the perception data in the miniaturized calibration field and the prior closed-loop constraint of the multi-lidar calibration, and optimize and solve to obtain the external parameter calibration parameters of the lidar and millimeter-wave radar. .

[0011] The present invention also provides a multi-group radar extrinsic parameter calibration system for an unmanned mining vehicle, the system comprising:

[0012] Miniaturized calibration field, consisting of five calibration plates with a specific spatial layout;

[0013] Point cloud acquisition unit, used to obtain the perception data of the laser radar and millimeter wave radar in the calibration field;

[0014] A feature extraction unit, used to process the perception data to extract the calibration plate boundary parameters, plane parameters and centroid information;

[0015] and a parameter optimization unit for optimizing and solving the radar extrinsic calibration parameters based on the extracted geometric constraint relationships.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. The present invention first extracts the calibration plate features through point cloud segmentation technology, and then uses plane fitting technology to extract the calibration plate plane data and boundary data. On this basis, the constraint relationship between multiple laser radars on the calibration plate plane and boundary parameters is established, and the rotation matrix and translation vector of the centimeter-level error between multiple laser radars are obtained through optimization solution.

[0018] 2. The present invention extracts center point data by segmenting the calibration plate point cloud, and accurately locates the center point of the calibration plate in combination with the significant echo characteristics of the corner reflector in the millimeter-wave radar. By establishing the spatial constraint relationship between the center point of the calibration plate detected by the lidar and the corner reflector echo point detected by the millimeter-wave radar and the prior closed-loop constraint of the multi-lidar calibration, the external calibration parameters between the lidar and the millimeter-wave radar are optimized and solved. For the calibration of multiple lidars and millimeter waves, it is suitable for the calibration of multiple lidars and millimeter waves of unmanned mining vehicles, thereby improving efficiency, thereby solving the problems of poor calibration accuracy, insufficient global optimization and scalability in the existing technology of unmanned mining vehicles in rugged outdoor environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flow chart of the method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle according to the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Example 1:

[0022] In a specific embodiment, Figure 1 As shown, the present invention provides a method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle, comprising the following steps:

[0023] The first step is to design a new calibration plate frame with a foam rectangular calibration plate as the core and a corner reflector. At the same time, based on this calibration plate frame, a miniaturized calibration field is designed that can meet the external parameter calibration task of multiple laser radars and millimeter wave radars for unmanned mining vehicles.

[0024] First, a new calibration plate frame is designed with a foam rectangular calibration plate as the core and a corner reflector. The corner reflector is placed in the center of the calibration plate, and the calibration target is rotated 45°. The size of the calibration plate is set to 85cm*120cm.

[0025] Then install 5 calibration plates in the calibration field. With the center of calibration plate 1 as the origin, the coordinates of the other calibration plates are: calibration plate 2 (2.0, 1), calibration plate 3 (-3.5, 1.5), calibration plate 4 (-2.0, 1) and calibration plate 5 (3.5, 1.5). The coverage area of ​​the calibration field does not exceed 5m. 2 .

[0026] The second step is to record the LiDAR data of the miniaturized calibration field to obtain the LiDAR point cloud data in the calibration field, and implement calibration target extraction and constraint extraction based on the calibration field point cloud data.

[0027] First, based on the lidar point cloud data collected in a miniaturized calibration field, taking lidar A as an example, a preliminary filtering is first performed through spatial position constraints. Then, the filtered target area point cloud is segmented using the DBSCAN algorithm, and each point cloud cluster is plane fitted using RANSAC. Point cloud clusters whose projection scales in three orthogonal directions do not exceed 1.5 times the theoretical maximum size of the calibration plate are determined to be valid calibration plate point clouds.

[0028] Secondly, based on the calibration plate point cloud collected by the lidar, the calibration plate point cloud data will be numbered according to the calibration field design plan. The number range is First, the RANSAC algorithm is used to perform plane projection processing on the calibration plate point cloud numbered i. Then, the boundary extraction operation is performed on the projected point cloud, and the boundary point cloud extraction is realized by the convex hull algorithm in the point cloud library PCL.

[0029] Then, for the extracted boundary point cloud, the curvature calculation method based on LOAM is used to perform feature quantification analysis:

[0030]

[0031] In the formula, c represents the curvature value of the boundary point cloud, and S represents the number of neighboring points participating in the single-point curvature calculation. The value is 4, which means that two adjacent points before and after the target point are selected. and They represent the distance measurement values ​​between the target point and the neighboring points respectively, perform descending sorting on the curvature values ​​of the boundary point cloud, select the four point clouds with the largest curvature as the four vertices of the rectangular calibration plate, and based on the spatial position information of these four vertices, divide the boundary point cloud of the calibration plate into four boundary line segments: upper left, lower left, upper right, and lower right. At the same time, obtain the internal plane point cloud after RANSAC projection.

[0032] Finally, for the five rectangular calibration plates, the RANSAC algorithm is used to fit a three-dimensional straight line to the extracted upper left, lower left, upper right and lower right boundary point clouds. The boundary parameters of the three-dimensional straight line are the point and direction vector. At the same time, three non-collinear points are randomly sampled from the plane point cloud of the calibration plate, two direction vectors are calculated and the plane normal vector is obtained by cross product. The direction of the normal vector is adaptively adjusted and normalized according to the position characteristics of the point cloud. Finally, the four parameters (nx, ny, nz, d) that conform to the point normal plane equation are output, where (nx, ny, nz) are the normalized normal vector components and d represents the distance from the plane to the origin of the coordinate system.

[0033] The third step is to establish the constraint relationship between each laser radar and the miniaturized calibration field point cloud data on the calibration target, and optimize and solve the rotation matrix With translation vector .

[0034] First, based on the above, the number is The four boundary parameters of upper left, lower left, upper right and lower right and the internal plane parameters extracted from the rectangular calibration plate are used to construct the geometric constraint relationship between the lidar A and B. The corresponding relationship between the boundary segments , derive the rotation matrix between lidars and translation vectors The two constraint formulas (1) and (2) are:

[0035] (1)

[0036] (2)

[0037] In the above formula, Represents the centroid position of boundary j on the calibration plate numbered i collected by the main lidar A, Represents the centroid position of boundary j on the calibration plate numbered i collected by the main lidar B; is the direction vector of the jth boundary of the i-th calibration plate of the lidar A. There are five calibration plate planes, and each plane has four boundaries; is the direction vector of the jth boundary of the i-th calibration plate of lidar B. There are five calibration plate planes, and each plane has four boundaries; Three rows and one column Matrix, each calibration plate boundary can be a rotation matrix and translation vectors The two constraint formulas (1) and (2) for The corresponding relationship between the plane parameters of the rotation matrix is ​​derived and translation vectors The constraint formulas (3) and (4) are:

[0038] (3)

[0039] (4)

[0040] In the above formula Represents dot product, based on the corresponding relationship of plane parameters , is the i-th plane corresponding to the laser radar A, is the i-th plane corresponding to the laser radar B; Represents the plane centroid point cloud calibrated by the lidar B with serial number i; is the normal vector of the i-th calibration plate plane of the laser radar A, and there are five calibration plate planes in total; is the normal vector of the i-th calibration plate plane of the lidar B, there are five calibration plate planes in total; is the normal vector of the i-th calibration plate plane of the laser radar A, and there are five calibration plate planes; each calibration plate plane parameter can be a rotation matrix and translation vectors Provides a single constraint.

[0041] Then, based on the subsection, we get the upper left, lower left, upper right, lower right four boundary parameters and internal plane parameters extracted from the rectangular calibration plate numbered i, and use formulas (1) and (3) to obtain , is calculated by minimizing the following cost function (5) :

[0042] (5)

[0043] Closed-form solution, defined as:

[0044]

[0045]

[0046] Assumptions The singular value decomposition (SVD) of , calculated .

[0047] Then give Estimates, Solved by constraint formulas (2) and (4), the points on the boundary point cloud are much smaller than the points on the plane point cloud. In order to avoid deviation, the plane centroid is used. To estimate , we can get formula (6) through formula (4):

[0048] (6)

[0049] Stacking the constraint equations of the translation vector of the five calibration targets, we get formula (7):

[0050] (7)

[0051] The estimation of is regarded as a least squares problem, and the estimation model is , , where each calibration target provides 1 surface constraint, defining , Indicates a given , the model predicts The deviation from the actual observed b; is a matrix composed of five normal vectors, b is five A matrix of five rows and one column; is a real matrix with five rows and three columns; the translation vector The estimation problem of is expressed as formula (8):

[0052] (8)

[0053] Formula (8) is the standard form of the least squares problem, which aims to find the optimal , so that the residual sum of squares is minimized, is the transpose of the matrix.

[0054] Solve by least squares method:

[0055]

[0056] Finally, after obtaining the initial estimated rotation matrix and translation vector, we jointly optimize formula (9) by minimizing the following cost function:

[0057] (9)

[0058] Formula (9) is a least squares problem. The Levenberg-Marquardt method is used to solve this problem. The error is minimized by iteratively optimizing the objective function to solve the rotation matrix between the lidars. and translation vectors .

[0059] The fourth step is to record the laser radar and millimeter wave radar data of the miniaturized calibration field, obtain the laser radar point cloud data and millimeter wave radar perception data of the calibration field, and extract the centroid information of the laser radar and millimeter wave radar calibration targets based on the calibration field perception data.

[0060] The reflected RCS value of the calibration target is in the range of 10-20dBsm. Initial filtering is first performed based on spatial position constraints to eliminate millimeter-wave detection data in non-calibration areas. Specifically, the data information is filtered out using appropriate x and y values. The millimeter-wave detection data is then further filtered by applying a filter with a value between 10 and 25dBsm. At the same time, only the calibration target points and some additional millimeter-wave detection data in the background remain. The approximate position of each calibration target is known through the miniaturized calibration field design scheme.

[0061] Then, based on the original point cloud data collected by the unmanned mining vehicle lidar, preliminary filtering is first performed through spatial position constraints to eliminate the interference point cloud of the ground point cloud and the non-calibrated area. Subsequently, the filtered target area point cloud is segmented using the DBSCAN density clustering algorithm, and each point cloud cluster is plane fitted using the RANSAC algorithm. Point cloud clusters with a point-to-plane distance less than 0.06, a proportion of internal points exceeding 80%, and projections in three orthogonal directions not exceeding 1.5 times the theoretical size of the calibration plate are judged as valid calibration plate point clouds. Finally, the coordinates of the centroid of each calibration plate plane point cloud are calculated using the PCA algorithm.

[0062] Step 5: Establish the constraint relationship between the lidar and millimeter-wave radar on the perception data in the miniaturized calibration field and the prior closed-loop constraint of the multi-lidar calibration, and optimize and solve to obtain the external calibration parameters of the lidar and millimeter-wave radar. Specifically:

[0063] The three sensors are represented by LiDARs A and B and millimeter-wave radar R. Based on the miniaturized calibration field, through the above-mentioned millimeter-wave radar and LiDAR calibration target feature extraction process, LiDAR A provides 5 calibration target detection information as follows: , LiDAR B provides 5 calibration target detection information as , the trihedral corner reflector information detected by the millimeter wave radar R is , expressed in two-dimensional Euclidean coordinates, the variable Indicates whether the laser radar A detects the calibration target, so if the target is not found or discarded, , otherwise it is 1. Similarly, the laser radar B and the millimeter wave radar R have calibration target variable values and , the extrinsic calibration of LiDAR and millimeter-wave radar aims to estimate the relative rigid transformation , which projects the centroid of the calibration target in the millimeter-wave radar onto the coordinate frame of the lidar. The rigid transformation is represented by a 3×3 rotation matrix and translation vectors Composition, described as a 4×4 matrix of homogeneous coordinates, as follows:

[0064]

[0065] For the i-th calibration target position, the transformation error formula (10) between the lidars is the square of the Euclidean distance between the calibration target centroid in lidar B and the coordinate frame of lidar A:

[0066] (10)

[0067] If one of the sensors is a millimeter-wave radar R, a different error term is used, let Indicates the calibration target The error formula (11) between the millimeter-wave radar A and the lidar R is the square of the Euclidean distance between the center of mass of the calibration target in the millimeter-wave radar R and the coordinate frame of the lidar A:

[0068] (11)

[0069] function It is a mapping function that expands a two-dimensional point into a three-dimensional point. Since the millimeter-wave radar can only obtain plane information, the centroid coordinates of the calibration target detected by it are Considering the installation height difference between the laser radar and the millimeter-wave radar on the unmanned mining vehicle, the elevation information (z coordinate) of the corresponding matching calibration target detected by the laser radar is assigned to the two-dimensional point detected by the millimeter-wave radar. , thereby realizing its three-dimensional space expansion; is the transformation matrix from millimeter wave radar to lidar.

[0070] The attitude estimation of LiDAR A and millimeter-wave radar R can be expressed as an optimization problem to find the best solution for LiDAR A and millimeter-wave radar R. The optimal transformation that minimizes the total error formula between the calibration targets is obtained, and the optimal calibration parameters are found by minimizing the error formula (12):

[0071] (12)

[0072] The sensor calibration program is based on the unmanned mining vehicle hardware platform This is the transformation matrix between LiDAR A and LiDAR B, which can be calculated by the multi-LiDAR extrinsic calibration algorithm based on the miniaturized calibration field. Formulas (13) and (14) can be optimized using the Ceres library. The initial transformation matrix is ​​optimized by and The parameters are obtained, where Finally, the external calibration parameters of the laser radar A and the millimeter wave radar are optimized. :

[0073] (13)

[0074] (14)

[0075] Formula (13) is and The sum of It is the square of the Euclidean distance between the center of mass of the calibration target in the millimeter-wave radar R and the coordinate frame of the lidar B.

[0076] Example 2:

[0077] The present invention also provides a multi-group radar extrinsic parameter calibration system for an unmanned mining vehicle, comprising:

[0078] A miniaturized calibration field consisting of five calibration plates with a specific spatial layout;

[0079] A point cloud acquisition unit for acquiring sensing data from the laser radar and millimeter-wave radar within the calibration field;

[0080] A feature extraction unit for processing the sensing data to extract the calibration plate boundary parameters, plane parameters and centroid information;

[0081] and a parameter optimization unit for optimizing and solving the radar inter-external calibration parameters based on the extracted geometric constraint relationship.

[0082] The feature extraction unit includes:

[0083] Point cloud segmentation module, used to cluster the filtered lidar point cloud using the DBSCAN algorithm;

[0084] Plane fitting module, used to fit point cloud cluster planes using the RANSAC algorithm;

[0085] The boundary extraction module is used to extract the boundary of the projected point cloud using the PCL convex hull algorithm and calculate the boundary curvature based on the LOAM curvature formula;

[0086] The parameterization module is used to fit four boundary lines through RANSAC and calculate the normalized normal vector (nx, ny, nz) and the plane distance parameter d through the cross product of the plane point cloud;

[0087] Millimeter wave data processing module, used to extract corner reflector target points through spatial position filtering and RCS value filtering, and locate effective targets by combining calibration field design coordinates;

[0088] and a centroid mapping module for expanding millimeter-wave radar two-dimensional detection points into three-dimensional points.

[0089] The parameter optimization unit includes:

[0090] The multi-lidar calibration module is used to establish geometric constraints between lidars A and B, solve the initial rotation matrix through SVD decomposition, solve the translation vector through least squares, and finally use the Levenberg-Marquardt algorithm to jointly optimize the objective function;

[0091] The laser-millimeter wave joint calibration module is used to establish a spatial mapping relationship between the millimeter wave radar's two-dimensional point cloud and the lidar's three-dimensional point cloud, fuse multi-sensor detection constraints, and optimize the solution of the extrinsic calibration parameters between the lidar and millimeter wave radar.

[0092] In summary, the present invention is suitable for the calibration of multiple laser radars and millimeter waves for unmanned mining vehicles, improving efficiency and thus solving the problems of poor calibration accuracy, insufficient global optimization and scalability in rugged outdoor environments in the existing technology of unmanned mining vehicles.

[0093] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. The method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle is characterized by: The following steps are involved: S1. A novel calibration plate framework is designed with a foam rectangular calibration plate as the core and in conjunction with a corner reflector. This calibration plate framework can also be used to create a miniaturized calibration field that can meet the external parameter calibration tasks of multiple laser radars and millimeter-wave radars for unmanned mining vehicles. S2. Record the lidar data of the miniaturized calibration field to obtain the lidar point cloud data in the calibration field, and extract the calibration target and constraint based on the calibration field point cloud data; S3. Establish the constraint relationship between each laser radar and the miniaturized calibration field point cloud data on the calibration target, and optimize and solve the rotation matrix With translation vector ; S4. Recording laser radar and millimeter-wave radar data for the miniaturized calibration field to obtain laser radar point cloud data and millimeter-wave radar perception data of the calibration field, and extracting the centroid information of the laser radar and millimeter-wave radar calibration targets based on the calibration field perception data; S5. Establish the constraint relationship between the lidar and millimeter-wave radar on the perception data in the miniaturized calibration field and the prior closed-loop constraint of the multi-lidar calibration, and optimize and solve to obtain the external parameter calibration parameters of the lidar and millimeter-wave radar. .

2. The method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle according to claim 1, characterized in that: The specific operation steps of step S1 are: S11. Design a new calibration plate frame with a foam rectangular calibration plate as the core and a corner reflector. The corner reflector is placed in the center of the calibration plate, the calibration target is placed at a 45° rotation angle, and the size of the calibration plate is set to 85cm*120cm. S12. Install five calibration plates in the calibration field. Take the center of calibration plate 1 as the origin. The coordinates of the other calibration plates are: calibration plate 2 (2.0, 1), calibration plate 3 (-3.5, 1.5), calibration plate 4 (-2.0, 1) and calibration plate 5 (3.5, 1.5). The coverage area of ​​the calibration field does not exceed 5m. 2 .

3. The method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle according to claim 2, characterized in that: The specific operation steps of step S2 are: S21. Based on the lidar point cloud data collected from the miniaturized calibration field, the data is first preliminarily filtered using spatial position constraints. The filtered target area point cloud is then segmented using the DBSCAN algorithm. RANSAC is used to perform plane fitting on each point cloud cluster. Point cloud clusters whose projection scales in three orthogonal directions do not exceed 1.5 times the theoretical maximum size of the calibration plate are determined to be valid calibration plate point clouds. S22, based on the calibration plate point cloud collected by the laser radar, the calibration plate point cloud data will be numbered according to the calibration field design plan, and the numbering range is First, the RANSAC algorithm is used to perform plane projection processing on the calibration plate point cloud numbered i. Then, the boundary extraction operation is performed on the projected point cloud, and the boundary point cloud extraction is realized by the convex hull algorithm in the point cloud library PCL; S23. For the extracted boundary point cloud, a curvature calculation method based on LOAM is used to perform feature quantitative analysis: ， In the formula, c represents the curvature value of the boundary point cloud, and S represents the number of neighboring points participating in the single-point curvature calculation. The value is 4, which means that two adjacent points before and after the target point are selected. and Represent the distance measurement value between the target point and the neighboring point respectively, perform descending sorting operation on the curvature value of the boundary point cloud, select the four point clouds with the largest curvature as the four vertices of the rectangular calibration plate, and based on the spatial position information of these four vertices, divide the boundary point cloud of the calibration plate into four boundary line segments: upper left, lower left, upper right, and lower right. At the same time, obtain the internal plane point cloud after RANSAC projection; S24. For the five rectangular calibration plates, the RANSAC algorithm is used to fit a three-dimensional straight line to the four extracted boundary point clouds of upper left, lower left, upper right and lower right. The three-dimensional straight line boundary parameters are the point and direction vector. At the same time, three non-collinear points are randomly sampled from the plane point cloud of the calibration plate, two direction vectors are calculated and the plane normal vector is obtained by cross product. The normal vector direction is adaptively adjusted and normalized according to the position characteristics of the point cloud. Finally, four parameters (nx, ny, nz, d) that conform to the point normal plane equation are output, where (nx, ny, nz) are the normalized normal vector components and d represents the distance from the plane to the origin of the coordinate system.

4. The method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle according to claim 3, characterized in that: The specific operation steps of step S3 are: S31, based on step S2, obtain the number The four boundary parameters of upper left, lower left, upper right and lower right and the internal plane parameters extracted from the rectangular calibration plate are used to construct the geometric constraint relationship between the lidar A and b. The corresponding relationship between the boundary segments , derive the rotation matrix between lidars and translation vectors Two constraint formulas: ， ， In the above formula, Represents the centroid position of boundary j on the calibration plate numbered i collected by the main lidar A, Represents the centroid position of boundary j on the calibration plate numbered i collected by the main lidar B; is the direction vector of the jth boundary of the i-th calibration plate of the lidar A. There are five calibration plate planes, and each plane has four boundaries; is the direction vector of the jth boundary of the i-th calibration plate of lidar B. There are five calibration plate planes, and each plane has four boundaries; Three rows and one column Matrix; each calibration plate boundary can be a rotation matrix and translation vectors The two constraint formulas for the corresponding relationship of the plane parameters numbered i derive the rotation matrix and translation vectors The constraint formula is: ， ， In the above formula Represents dot product, based on the corresponding relationship of plane parameters , is the i-th plane corresponding to the laser radar A, is the i-th plane corresponding to the laser radar B; Represents the plane centroid point cloud calibrated by the lidar B with serial number i; is the normal vector of the i-th calibration plate plane of the laser radar A, and there are five calibration plate planes in total; is the normal vector of the i-th calibration plate plane of the lidar B, there are five calibration plate planes in total; is the normal vector of the i-th calibration plate plane of the laser radar A, and there are five calibration plate planes; each calibration plate plane parameter can be a rotation matrix and translation vectors Provide an independent constraint; S32, obtain the upper left, lower left, upper right, lower right four boundary parameters and internal plane parameters extracted by the rectangular calibration plate numbered i, and use the formula 、 To obtain , calculated by minimizing the following cost function formula : ， Closed-form solution, defined as: ， ， Assumptions The singular value decomposition of , calculated ; S33, given Estimates, By constraint formula and To solve this problem, the points on the boundary point cloud are much smaller than the points on the plane point cloud. In order to avoid deviation, the plane centroid is used. To estimate , through The following formula is obtained: ， By stacking the constraint equations of the five calibration targets about the translation vector, we get the following formula: ， The estimation of is regarded as a least squares problem, and the estimation model is , , where each calibration target provides 1 surface constraint, defining , the translation vector The estimation problem of is expressed as the following formula: , the above formula is the standard form of the least squares problem, whose purpose is to find the optimal , so that the residual sum of squares is minimized, is the transpose of the matrix; Solve by least squares method: ； S34. After obtaining the initial estimated rotation matrix and translation vector, the formula is jointly optimized by minimizing the following cost function: ， The above formula is a least squares problem. The Levenberg-Marquardt method is used to solve this problem. The error is minimized by iteratively optimizing the objective function to solve the rotation matrix between the lidars. and translation vectors .

5. The method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle according to claim 4, characterized in that: The specific operation steps of step S4 are: S41. The reflected RCS value of the calibration target is in the range of 10-20dBsm. First, preliminary filtering is performed based on spatial position constraints to eliminate millimeter wave detection data in non-calibration areas. The approximate position of each calibration target is obtained through the miniaturized calibration field design scheme. S42. Based on the original point cloud data collected by the unmanned mining vehicle lidar, preliminary filtering is first performed through spatial position constraints to eliminate the interference point cloud of the ground point cloud and the non-calibrated area. Then, the filtered target area point cloud is segmented using the DBSCAN density clustering algorithm, and each point cloud cluster is plane fitted using the RANSAC algorithm. Point cloud clusters with a point-to-plane distance less than 0.06, an internal point ratio of more than 80%, and projections in three orthogonal directions not exceeding 1.5 times the theoretical size of the calibration plate are determined to be valid calibration plate point clouds. Finally, the coordinates of the centroid of each calibration plate plane point cloud are calculated using the PCA algorithm.

6. The method for calibrating multiple radar extrinsic parameters of an unmanned mining vehicle according to claim 5, characterized in that: The specific operation steps of step S5 are: S51. The three sensors are represented by laser radars A and B and millimeter wave radar R. Based on the miniaturized calibration field, through the above-mentioned millimeter wave radar and laser radar calibration target feature extraction process, the laser radar A provides 5 calibration target detection information as follows: , LiDAR B provides 5 calibration target detection information as , the trihedral corner reflector information detected by the millimeter wave radar R is , expressed in two-dimensional Euclidean coordinates, the variable Indicates whether the laser radar A detects the calibration target, so if the target is not found or discarded, Otherwise, it is 1, and there is a calibration target variable value between the laser radar B and the millimeter wave radar R and , the extrinsic calibration of LiDAR and millimeter-wave radar aims to estimate the relative rigid transformation , which projects the centroid of the calibration target in the millimeter-wave radar onto the coordinate frame of the lidar. The rigid transformation is represented by a 3×3 rotation matrix and translation vectors Composition, described as a 4×4 matrix of homogeneous coordinates, as follows: ， S52. For the i-th calibration target position, the transformation error formula between the lidars is as follows, which is the square of the Euclidean distance between the calibration target centroid in lidar B and the coordinate frame of lidar A: ， If one of the sensors is a millimeter-wave radar R, a different error term is used, let Indicates the calibration target The error formula between the millimeter-wave radar A and the lidar R is as follows, which is the square of the Euclidean distance between the centroid of the calibration target in the millimeter-wave radar R and the coordinate frame of the lidar A: ， function It is a mapping function that expands a two-dimensional point into a three-dimensional point. Since the millimeter-wave radar can only obtain plane information, the centroid coordinates of the calibration target detected by it are Considering the installation height difference between the laser radar and the millimeter-wave radar on the unmanned mining vehicle, the elevation information of the corresponding matching calibration target detected by the laser radar is assigned to the two-dimensional point detected by the millimeter-wave radar. , thereby realizing its three-dimensional space expansion; is the transformation matrix from millimeter wave radar to lidar; S53, the attitude estimation of laser radar A and millimeter wave radar R is expressed as an optimization problem to find the The optimal transformation that minimizes the total error formula between the calibration targets, and by minimizing the error formula To find the optimal calibration parameters: ； S54, sensor calibration program is based on the unmanned mining vehicle hardware platform This is the transformation matrix between LiDAR A and LiDAR B, which can be calculated by the multi-LiDAR extrinsic calibration algorithm based on a miniaturized calibration field: ， ； The above formula can be optimized using the Ceres library. The initial transformation matrix is ​​optimized by and The parameters are obtained, where Finally, the external calibration parameters of the laser radar A and the millimeter wave radar are optimized. ;formula for and The sum of It is the square of the Euclidean distance between the center of mass of the calibration target in the millimeter-wave radar R and the coordinate frame of the lidar B.

7. An unmanned mine vehicle multi-group radar extrinsic parameter calibration system, using the unmanned mine vehicle multi-group radar extrinsic parameter calibration method according to any one of claims 1 to 6, characterized in that: The system comprises: Miniaturized calibration field, consisting of five calibration plates with a specific spatial layout; Point cloud acquisition unit, used to obtain the perception data of the laser radar and millimeter wave radar in the calibration field; A feature extraction unit, used to process the perception data to extract the calibration plate boundary parameters, plane parameters and centroid information; and a parameter optimization unit for optimizing and solving the radar extrinsic calibration parameters based on the extracted geometric constraint relationships.

8. The unmanned mining vehicle multi-group radar extrinsic parameter calibration system according to claim 7, characterized in that: The feature extraction unit includes: Point cloud segmentation module, used to cluster the filtered lidar point cloud using the DBSCAN algorithm; Plane fitting module, used to fit point cloud cluster planes using the RANSAC algorithm; The boundary extraction module is used to extract the boundary of the projected point cloud using the PCL convex hull algorithm and calculate the boundary curvature based on the LOAM curvature formula; The parameterization module is used to fit four boundary lines through RANSAC and calculate the normalized normal vector (nx, ny, nz) and the plane distance parameter d through the cross product of the plane point cloud; Millimeter wave data processing module, used to extract corner reflector target points through spatial position filtering and RCS value filtering, and locate effective targets by combining calibration field design coordinates; and a centroid mapping module for expanding millimeter-wave radar two-dimensional detection points into three-dimensional points.

9. The unmanned mining vehicle multi-group radar extrinsic parameter calibration system according to claim 7, characterized in that: The parameter optimization unit includes: The multi-lidar calibration module is used to establish geometric constraints between lidars A and B, solve the initial rotation matrix through singular value decomposition, solve the translation vector through least squares, and finally use the Levenberg-Marquardt algorithm to jointly optimize the objective function; The laser-millimeter wave joint calibration module is used to establish a spatial mapping relationship between the millimeter wave radar's two-dimensional point cloud and the lidar's three-dimensional point cloud, fuse multi-sensor detection constraints, and optimize the solution of the extrinsic calibration parameters between the lidar and millimeter wave radar.

Images