Calibration method based on multiple radars and cameras

Abstract

The invention discloses a calibration method based on multiple radars and cameras, and the method comprises the steps: fusing point clouds of multiple radars under a static condition, carrying out the accumulation, obtaining the point clouds of a scene after accumulation, and recording the image information of the scene through a forward camera; performing ground removal on the accumulated point clouds of the scene based on a random consistency sampling method to obtain first point clouds, and by setting a preset height, filtering the point clouds of which the heights are greater than the preset height in the first point clouds to obtain second point clouds; and performing external parameter calibration on the plurality of radars and the forward camera based on the second point cloud and the image information of the scene recorded by the forward camera. According to the invention, data of the laser radar and the camera are fused, and the defects of a single sensor in the aspect of environmental perception are overcome; the requirement for time synchronization of the radar and the camera is low, and the problem that calibration cannot be carried out due to the fact that radar real-time frames and camera registration time are not synchronous is solved; by removing the ground point cloud, the problem of calibration of sparse and large-wide-angle point cloud and external parameters of the camera is solved.

Description

Technical Field

[0001] The present invention relates to the field of simultaneous localization and mapping of intelligent robots. More specifically, it relates to a calibration method based on multiple radars and cameras. Background Art

[0002] With the continuous development of technology, robots and autonomous systems are increasingly widely used in various fields, such as home service, industrial manufacturing, autonomous driving, etc. For these systems to achieve autonomous navigation and complete tasks in unknown or complex environments, they need to have accurate environmental perception and positioning capabilities. The Simultaneous Localization and Mapping (SLAM) technology can meet this requirement. The classic SLAM technology uses pure lidar for mapping. However, with the continuous progress of technology, the demand for cameras is becoming increasingly obvious. The fused mapping of cameras and radars has significant advantages in improving mapping accuracy, enhancing environmental adaptability, improving target detection capabilities, enhancing the quality of 3D maps, and supporting semantic map construction. This fusion technology is of great significance for the development of fields such as autonomous driving and intelligent robots.

[0003] However, the calibration of lidar and cameras is an essential step in the fused SLAM mapping of lidar and cameras. The lidar and cameras respectively obtain 3D point cloud data and 2D image data, and there are natural differences between them. To fuse these two types of data, it is necessary to first determine their spatial relationship, that is, to perform calibration. Only through accurate calibration can it be ensured that the fused data is consistent in space, providing a reliable basis for subsequent tasks such as environmental perception and target detection. In the fully automatic calibration method, the point cloud of the MID-360 radar is relatively sparse and the scanning angle is too large. It is very difficult to obtain correct calibration parameters using the MID-360 radar. Although it is possible to find outdoor scenes with clear textures, obvious corner points or edge features, and a clean background, the success rate is still relatively low and the environmental requirements are high. In the semi-automatic calibration method, real-time calibration is performed using single-frame radar data and camera data. First, the requirement for time synchronization is relatively high. Secondly, for the relatively sparse point cloud of MID-360, the edge of the calibration board is not clear in the point cloud of the radar during calibration, and effective calibration parameters cannot be obtained through the calibration process.

[0004] The lidar can obtain high-precision and discrete 3D point cloud data on the target surface, but the data is relatively sparse; the camera can obtain high-resolution and high-quality 2D visual information, but is easily affected by factors such as light and shadow. Summary of the Invention

[0005] The present invention provides a calibration method based on multiple radars and cameras to solve at least one of the problems existing in the prior art.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention provides a calibration method based on multiple radars and cameras, and the method includes

[0008] Under static conditions, fuse the point clouds of multiple radars and accumulate them to obtain the point cloud of the accumulated scene, and record the image information of the scene through the forward camera;

[0009] Remove the ground from the point cloud of the accumulated scene based on the method of random sample consensus to obtain the first point cloud, and filter out the point cloud with a height greater than the preset height in the first point cloud by setting a preset height to obtain the second point cloud;

[0010] Perform external parameter calibration on multiple radars and the forward camera based on the second point cloud and the image information of the scene recorded by the forward camera.

[0011] Optionally, the method of removing the ground from the point cloud of the accumulated scene based on the method of random sample consensus includes

[0012] Select the first seed point set according to the preset lowest point representative;

[0013] Set a plane model according to the first seed point set;

[0014] Set a plane distance threshold, and optimize the plane model based on the plane distance threshold.

[0015] Optionally, the selection of the first seed point set according to the preset lowest point representative includes

[0016] Select the points lower than the preset lowest point in the point cloud of the accumulated scene, and calculate their average value;

[0017] Select the points with a height less than the average value in the point cloud of the scene to form the first seed point set.

[0018] Optionally, the optimization of the plane model based on the plane distance threshold includes

[0019] Calculate the distance from all points in the point cloud of the accumulated scene to the orthogonal projection of the plane according to the plane model, and judge whether to determine the point as a ground point according to the distance: if the distance is less than the plane distance threshold, the point is a ground point; if the distance is greater than or equal to the plane threshold, the point is a non-ground point;

[0020] Use the ground points as the second seed point set, and perform iterative optimization until the number of iterations is reached and the optimization is stopped.

[0021] Optionally, the recording of the image information of the scene through the forward camera includes

[0022] Record the image information of the scene containing the calibration board through the front camera.

[0023] Optionally, the external parameter calibration of multiple radars and the front camera includes

[0024] Obtain the vertices of the calibration board in the radar coordinate system and the corner points of the calibration board in the camera image;

[0025] Calculate the correspondence between the vertices of the calibration board in the radar coordinate system and the corner points of the calibration board in the image, and perform external parameter calibration.

[0026] Optionally, before the external parameter calibration of multiple radars and the front camera, it also includes

[0027] Calibrate the internal parameters of the front camera through the toolkit.

[0028] Optionally, the calculation of the correspondence between the vertices of the calibration board in the radar coordinate system and the corner points of the calibration board in the image includes

[0029] Project the target in the radar coordinate system into the image coordinate system, and the projection relationship formula is

[0030]

[0031] Among them, u', v', and w' respectively represent the coordinates of the points in the radar coordinate system, represents the internal parameters of the camera, represents the known determined matrix, represents the rotation matrix from the radar coordinate system to the image coordinate system, represents the translation matrix from the radar coordinate system to the image coordinate system, x i 、y i 、z i respectively represent the coordinate values of the points in the image coordinate system.

[0032] Optionally, the method further includes obtaining the mapping relationship between the radar coordinate system and the image coordinate system through the PnP method for external parameter calibration.

[0033] Optionally, the method further includes verifying the accuracy of the external parameter calibration through the calibration board at a known position.

[0034] The beneficial effects of the present invention are as follows:

[0035] The present invention combines a fully automatic calibration method and a semi-automatic calibration method. Through the calibration between a lidar and a camera, the data of both are fused to ensure the consistency and accuracy of the fused data, thereby overcoming the shortcomings of a single sensor in environmental perception and improving the robustness of the system. The present invention has relatively low requirements for the time synchronization between the lidar and the forward camera, overcoming the problem that calibration cannot be performed due to time asynchrony during the registration of the real-time frame of the radar and the camera. By removing the ground point cloud and the point cloud outside the preset height, a large number of interfering point clouds during calibration projection are reduced, solving the problem of external parameter calibration between sparse, wide-angle point clouds and the camera. This solution is also applicable to the external parameter calibration of other types of radars and cameras. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The following further elaborates on the specific embodiments of the present invention with reference to the accompanying drawings.

[0037] Figure 1 The flowchart showing the calibration method of multiple radars and cameras in the present invention is presented;

[0038] FIG. 2(a) shows a schematic diagram of the positions of four radars (two in the front and two in the back) installed on the robotic dog, and FIG. 2(b) shows a schematic diagram of the position of the forward camera installed on the robotic dog;

[0039] Figure 3 The schematic diagram showing the point cloud collected in real time by the four radars and a single-frame image of the forward camera in the present invention is presented;

[0040] Figure 4 The schematic diagram showing the point cloud image after accumulation in the present invention is presented;

[0041] Figure 5 The schematic diagram showing the accumulated point cloud after removing the ground and the roof of the 3D point cloud map using the method based on random sample consensus in the present invention is presented;

[0042] Figure 6 The schematic diagram showing the image of successfully extracting the calibration plate plane in the lidar point cloud in the present invention is presented;

[0043] Figure 7 The schematic diagram showing the effect of external parameter calibration of four radars and one forward camera in the present invention is presented. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To more clearly illustrate the present invention, the following further describes the present invention with reference to preferred embodiments and the accompanying drawings. Similar components in the drawings are denoted by the same reference numerals. Those skilled in the art should understand that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the protection scope of the present invention.

[0045] The present invention combines a semi-automatic calibration method and a full-automatic calibration method, and proposes a calibration method based on multiple radars and cameras. It improves the existing calibration method for lidar and cameras for multiple MID-360 lidars, fuses the data of the two, and ensures the consistency and accuracy of the fused data, thereby overcoming the shortcomings of a single sensor in environmental perception and solving the problem of external parameter calibration for multiple MID-360 radars and cameras.

[0046] As Figure 1 shown, this method includes

[0047] Fusing the point clouds accumulated by multiple radars under static conditions to obtain the accumulated scene point cloud, and recording the image information of the scene through the forward camera;

[0048] Recording the image information of the scene through the forward camera includes recording the image information of the scene containing the calibration board through the forward camera;

[0049] Fusing the point clouds of multiple radars into a large point cloud, and using the fused large point cloud for temporal accumulation, and the accumulated large point cloud is denser.

[0050] During the accumulation process, a large amount of point clouds accumulate at the roof and ground positions, causing interference to the external parameter calibration. The method of random sample consensus is used to remove the ground from the accumulated scene point cloud to obtain the first point cloud; by setting a preset height, the point clouds in the first point cloud with a height greater than the preset height are filtered out to obtain the second point cloud;

[0051] Selecting the first seed point set according to the preset lowest point representative: selecting the points in the accumulated scene point cloud that are lower than the preset lowest point, and calculating their average value; selecting the points in the scene point cloud with a height less than the average value to form the first seed point set;

[0052] Setting a plane model according to the first seed point set;

[0053] Setting a plane distance threshold, and optimizing the plane model based on the plane distance threshold.

[0054] Calculating the distance from all points in the accumulated scene point cloud to the orthogonal projection of this plane according to the plane model, and judging whether to determine this point as a ground point according to the distance: if the distance is less than the plane distance threshold, this point is a ground point; if the distance is greater than or equal to the plane threshold, this point is a non-ground point; using the ground points as the second seed point set, and iteratively optimizing until the number of iterations is reached, stopping the optimization, and realizing the removal of the ground points in the accumulated radar point cloud to obtain the first point cloud.

[0055] Based on the method of random consistency sampling, the roof point cloud is further filtered. Specifically, by setting a preset height, the point cloud in the first point cloud with a height greater than the preset height is filtered to obtain the second point cloud.

[0056] For the calibration of the internal parameters of the forward camera, the tool kit for calibrating the camera in the fourth long-term support version (ROS Melodic) of the Robot Operating System (ROS) is used to calibrate the internal parameters of the forward camera on the robotic dog, which serves as the input parameter during the calibration of the external parameters.

[0057] Calibrate the external parameters of multiple radars and the forward camera.

[0058] Obtain the vertices of the checkerboard calibration board in the radar coordinate system and the corner points of the checkerboard calibration board in the camera image respectively.

[0059] Solve the mapping relationship between the vertices of the checkerboard calibration board in the radar coordinate system and the corner points of the checkerboard calibration board in the camera image to achieve the calibration of the external parameters.

[0060] Use other calibration boards or feature points with known positions to verify the accuracy of the calibration; the solved external parameters can be applied to the subsequent data fusion of the lidar and the camera.

[0061] In a specific embodiment, a robotic dog is used as the carrier of the lidar and the forward camera. Four lidars and one forward camera are set for the calibration of the external parameters. The carrier of the lidar and the forward camera can also be a fixed platform or an aircraft. Figure 2(a) is a schematic diagram of setting four lidars on the robotic dog, and Figure 2(b) is a schematic diagram of the position of the forward camera. Specifically,

[0062] The scene point clouds accumulated by the four lidars are stitched through the angles and positions of the four lidars installed on the robotic dog. The point clouds of multiple lidars are fused into a large point cloud, and a complete scene point cloud is generated and published through the Ros topic. Figure 3 Figure [ID] is a schematic diagram of the single-frame image of the four radar point clouds and the camera collected in real time. Figure 4 Figure [ID] is a schematic diagram of the point cloud image after fusing and accumulating the four scene point clouds.

[0063] Prepare a calibration board, measure the length, width and thickness of the calibration board, which serves as the input parameter during the calibration of the external parameters of the four lidars and the forward camera. Its shape is generally quadrilateral. In this embodiment, a checkerboard calibration board is used, with a length of 54.1 cm, a width of 34.3 cm, and a thickness of 0.6 cm.

[0064] Keep the robotic dog stationary, accumulate point clouds through lidar, and collect image data through the front camera. Among them, the accumulation time of the lidar point clouds is not specifically limited, and it is sufficient to accumulate until the relative position between the calibration board and the environment can be clearly seen. In this embodiment, the accumulation time is set to 20 seconds.

[0065] During the accumulation process, a large number of point clouds in the scene accumulate at the roof and ground positions, interfering with the external parameter calibration. The method of random sample consensus is used to remove the ground from the accumulated scene point clouds to obtain the first point cloud. By setting a preset height, the point clouds in the first point cloud that are higher than the preset height are filtered to obtain the second point cloud. The specific method is

[0066] In the scene point clouds accumulated by the lidar, a first seed point set is selected according to the preset lowest point, and the lowest point representative is set. In the accumulated scene point clouds, the point clouds lower than the lowest point representative are selected, and the average value of the point cloud heights is calculated. All the points in the scene point clouds with heights lower than this average value are used as seed points to form the first seed point set; in this embodiment, a0x + b0y + c0z + d0 = 0 is used as the plane estimation model, that is, n T x = -d0, where x, y, and z respectively represent the coordinates of the point cloud, and a0, b0, c0, and d0 all represent constants, and the vector n = [a0, b0, c0] T , the vector x = [x, y, z] T , T represents the transpose. In this embodiment, S ∈ R 3 is used as the initial point set, where R 3 represents 3D, and its covariance matrix is where Cov represents the covariance matrix, S represents the first seed point set, represents the mean value of all seed points. The normal vector n perpendicular to the plane in the plane model represents the direction with the smallest variance, and n is obtained by calculating the singular vector of the smallest singular value. The constant d0 can be directly obtained by substituting the average value of the seed point set directly.

[0067] After determining the plane model, a plane distance threshold is set to optimize the plane model. Further calculate the distance from each point in the lidar to the orthogonal projection of the plane. Compare the calculated distance with the preset plane distance threshold: if the calculated distance is less than the plane distance threshold, then determine that the point is a ground point; if the calculated distance is greater than the plane distance threshold, then determine that the point is a non-ground point. The ground points in the scene point clouds are used as the second seed point set for the next iteration, and iterative optimization is performed until the number of iterations reaches the preset number of iterations, and then the optimization is stopped to achieve the removal of the ground point clouds.

[0068] Based on the method of random consensus sampling, further filter out the points that are too high (mainly the roof here) in the scene point cloud accumulated by the lidar. By setting the height, the scene point cloud is cropped according to the height value. In this embodiment, the height value is set to 2.6m, and the point cloud with a height value greater than 2.6m is filtered out to achieve the removal of the roof point cloud. Figure 5 It is a schematic diagram of the point cloud after ground removal and roof removal by the method of random consensus sampling.

[0069] Calibrate the internal parameters of the forward camera as the input parameters during the external parameter calibration; use the toolbox for calibrating the camera in the ROS Melodic distribution to calibrate the internal parameters of the forward camera on the robot dog.

[0070] Calibrate the external parameters of the four lidars and the forward camera on the robot dog. Specifically,

[0071] Obtain the vertices of the checkerboard calibration board in the radar coordinate system. Figure 6 It is a schematic diagram of extracting the planar image of the calibration board from the lidar point cloud. Use PC to represent the radar point cloud of the target, and X i represents the 3D point. The radar point cloud can be expressed as N represents the number of target points. In this embodiment, a quadrilateral checkerboard calibration board is used, and N is taken as 4.

[0072] Define the cost function as

[0073]

[0074] Among them, represents the mapping from the camera coordinate system to the radar coordinate system. respectively represent the point cloud coordinates in the radar coordinate system, ε represents the parameter to be optimized, and d represents the distance between the checkerboard calibration board and the origin of the fused radar coordinate system.

[0075]

[0076] Among them, λ ∈ R, a ≥ 0. λ and a have no practical meaning. λ corresponds to and in the cost function, and a corresponds to ε and

[0077] Manually click to obtain the corner points of the checkerboard calibration board in the image, and click the four corner points in the image in clockwise or counterclockwise order as the points corresponding to the vertices of the calibration board in the radar coordinate system.

[0078] Solve the corresponding relationship between the corner points of the checkerboard calibration board in the image and the radar vertices.

[0079] In this embodiment, Y iC Represents the corner points of the checkerboard calibration board in the camera image, Represents the corner points in the camera image. The radar target is projected onto the image coordinates to obtain the correspondence between the 3D point cloud in the lidar and the corner points of the checkerboard calibration board in the camera image. The conversion relationship between the radar coordinate system and the camera coordinate system is

[0080]

[0081] Wherein, Represents the coordinates corresponding to the points in the radar coordinate system, Represents the camera internal parameters, Represents the rotation matrix from the radar coordinate system to the image coordinate system, Represents the translation matrix from the radar coordinate system to the image coordinate system, Represents the transformation matrix, x i 、y i 、z i Respectively represent the coordinates of the corresponding points in the camera coordinate system.

[0082] Define the mapping relationship from the radar coordinate system to the image coordinate system as Obtain multiple pairs of 3D and 2D matching points through the PnP algorithm (Perspective-n-Point). In the case of known or unknown camera internal parameters, an algorithm for solving the camera external parameters by minimizing the reprojection error. In this embodiment, the mapping relationship between the lidar coordinate system and the image coordinate system is obtained through the PnP algorithm, and the formula is

[0083]

[0084] Wherein, Y i C Represents the corner points of the checkerboard calibration board in the camera image, Y i C ∈R 2 , n represents the number of poses of the target, Y i L Represents the mapping relationship from the radar coordinate system to the image coordinate system, Represents the optimal rotation matrix from the radar coordinate system to the image coordinate system, Represents the optimal translation matrix from the radar coordinate system to the image coordinate system, |||| represents the norm operation.

[0085] Define Wherein, Represents the best point cloud point in the radar coordinate system corresponding to the corner points of the image checkerboard, Represents the average value of i, Represents the best conversion from the radar to the target.

[0086] Solving the transformation matrix through the PnP algorithm Among them, is the rotation matrix from the radar coordinate system to the image coordinate system, representing the relative attitude between the two coordinate systems, is the translation matrix from the radar coordinate system to the image coordinate system, representing the relative position between the two coordinate systems. This transformation matrix is the external parameter of the four lidars and the forward camera on the robotic dog, Figure 7 is the calibration effect diagram of this embodiment.

[0087] Other calibration boards or feature points with known positions can be used to verify the calibration accuracy; apply the solved mapping relationship between the radar coordinate system and the image coordinate system to subsequent lidar and camera data fusion, such as in SLAM mapping for multi-lidar and camera fusion.

[0088] This embodiment proposes a solution for the external parameter calibration of four MID-360 radars and a forward camera, solving the problem of external parameter calibration between sparse, wide-angle point clouds and the camera. This solution is also applicable to the calibration of other types of radars and cameras.

[0089] This method combines the ideas of two classic calibration algorithms, full-automatic calibration and semi-automatic calibration. It uses the accumulated scene point cloud after filtering and other processing for calibration with the camera image, improving the semi-automatic calibration algorithm. This invention has a low requirement for the time synchronization between the radar and the camera, solving the problem that the semi-automatic calibration algorithm cannot be calibrated due to time asynchronization when using the real-time frame of the radar and the camera registration, so as to obtain the corresponding transformation matrix results of multiple radars and cameras; at the same time, this invention draws on the idea of accumulating point clouds in the full-automatic calibration method and overcomes the defect that the full-automatic calibration method has too high requirements for the scene.

[0090] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper" and "lower" is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention. Unless otherwise clearly specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0091] It should also be noted that in the description of the present invention, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent in such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the said element.

[0092] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or variations can be made on the basis of the above description. It is impossible to list all the implementation manners here. Any obvious changes or variations derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A calibration method based on multi-radar and camera, characterized in that, The method includes: Fusing the point clouds of multiple radars under static conditions and accumulating them to obtain the point cloud of the accumulated scene, and recording the image information of the scene through the front camera; Removing the ground from the point cloud of the accumulated scene based on the method of random sample consensus to obtain the first point cloud, and filtering out the point cloud with a height greater than the preset height in the first point cloud by setting a preset height to obtain the second point cloud; Calibrating the external parameters of multiple radars and the front camera based on the second point cloud and the image information of the scene recorded by the front camera.

2. The method according to claim 1, wherein The method of removing the ground from the accumulated scene point cloud based on random sample consensus includes Selecting the first seed point set according to the preset lowest point representative; Setting a plane model according to the first seed point set; Setting a plane distance threshold and optimizing the plane model based on the plane distance threshold.

3. The method according to claim 2, wherein The step of selecting the first seed point set according to the preset lowest point representative includes Selecting the points below the preset lowest point in the accumulated scene point cloud and calculating their average value; Selecting the points with a height less than the average value in the scene point cloud to form the first seed point set.

4. The method according to claim 2, characterized in that, The step of optimizing the plane model based on the plane distance threshold includes Calculating the distance from all points in the accumulated scene point cloud to the orthogonal projection of the plane according to the plane model, and judging whether the point is determined as a ground point according to the distance: if the distance is less than the plane distance threshold, the point is a ground point; if the distance is greater than or equal to the plane threshold, the point is a non-ground point; Taking the ground points as the second seed point set and iteratively optimizing until the iteration times are reached and the optimization is stopped.

5. The method according to claim 1, characterized in that, The step of recording the scene image information through the front camera includes Recording the image information of the scene including the calibration board through the front camera.

6. The method according to claim 5, wherein Calibrating the external parameters of multiple radars and the front camera includes Obtaining the vertices of the calibration board in the radar coordinate system and the corner points of the calibration board in the camera image; Calculating the corresponding relationship between the vertices of the calibration board in the radar coordinate system and the corner points of the calibration board in the image for external parameter calibration.

7. The method according to claim 1, wherein Before calibrating the external parameters of multiple radars and the front camera, it also includes Calibrating the internal parameters of the front camera through a tool kit.

8. The method according to claim 6, wherein The step of calculating the corresponding relationship between the vertices of the calibration board in the radar coordinate system and the corner points of the calibration board in the image includes Projecting the target in the radar coordinate system into the image coordinate system, and the projection relation formula is Among them, u', v', and w' respectively represent the coordinates of points in the radar coordinate system, represents the internal parameters of the camera, represents a known and determined matrix, represents the rotation matrix from the radar coordinate system to the image coordinate system, represents the translation matrix from the radar coordinate system to the image coordinate system, x i 、y i 、z i respectively represent the coordinate values of points in the image coordinate system.

9. The method according to claim 8, characterized in that, The method also includes obtaining the mapping relationship between the radar coordinate system and the image coordinate system through the PnP method for external parameter calibration.

10. The method according to claim 1, characterized in that The method also includes verifying the accuracy of the external parameter calibration through a calibration board at a known position.

Images