Method and system for spatial calibration of radar sensor, vehicle and computer program product
By building a matrix of equal point cloud data volumes and combining singular value decomposition method and ICP or Gauss Newton method, fast and efficient space calibration of radar sensors is achieved, solving the problem of complex, time-consuming calibration in the existing technology, and improving calibration quality .
Patent Information
- Application Number
- CN202411218548.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-12-06
AI Technical Summary
The spatial calibration methods of existing radar sensors are complex and time-consuming, requiring a large number of human optimization and iterative parameter adjustments, with low calibration quality, making it difficult to efficiently and accurately calibrate various types of radar sensors.
By constructing a point source matrix and point target matrix with equal point cloud data volumes, coarse registration is performed using singular value decomposition, and precise registration is performed by combining ICP algorithm or Gaussian Newton's method, iteratively updates the translation matrix and rotation matrix, and quickly and efficiently Realize spatial calibration of millimeter-wave radar and lidar.
It effectively reduces the number of iterations of the ICP algorithm or Gaussian Newton method, speeds up the convergence speed of the solution process, realizes fast, efficient and accurate spatial calibration of radar sensors, and reduces time cost and technical overhead.
Smart Images

Figure CN119087369A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of radar sensors, and in particular to a method for spatial calibration of a radar sensor, a system for spatial calibration of a radar sensor, a vehicle comprising the system according to the present application, and a computer program product for at least assisting in implementing the steps of the method according to the present application. Background Art
[0002] When a vehicle performs high-level assisted driving functions, it is usually necessary to fuse the perception data of various types of radar sensors, especially millimeter-wave radar and lidar, for decision-making and body control. The current front-fusion perception solution can combine the characteristics of lidar and millimeter-wave radar, such as the accuracy and density of lidar's point cloud and the Doppler information of millimeter-wave radar's point cloud, thereby giving the fused perception data more information content and improving the robustness of the fused perception data.
[0003] The front-fusion perception solution relies on the spatial calibration of each radar sensor. The traditional calibration method requires each radar sensor to be calibrated separately and then integrated. The calibration process is complicated and requires a lot of manual optimization and iterative parameter adjustment. It also has strict requirements on the calibration scene. This results in high time cost and technical overhead for radar sensor calibration, and low calibration quality.
[0004] Therefore, how to efficiently and accurately calibrate various types of radar sensors has become a technical problem that needs to be solved. Summary of the invention
[0005] The purpose of the present application is to provide a method for spatial calibration of a radar sensor, a system for spatial calibration of a radar sensor, a vehicle including the system according to the present application, and a computer program product, so as to at least partially solve the problems in the prior art.
[0006] According to a first aspect of the present application, a method for spatial calibration of a radar sensor is provided, the method comprising:
[0007] - Step S1: constructing a point source matrix B based on first point cloud data about at least three calibration objects collected by a first radar sensor, and constructing a point target matrix A based on second point cloud data about the at least three calibration objects collected by a second radar sensor, wherein the point source matrix B and the point target matrix A contain an equal amount of point cloud data;
[0008] - Step S2: Obtain the translation matrix T and rotation matrix R required to transform the point source matrix B into the point target matrix A by singular value decomposition method; and
[0009] -Step S3: Using the obtained translation matrix T and rotation matrix R as initial variables, iteratively update the translation matrix T and the rotation matrix R.
[0010] The core concept of the present application includes: constructing a point source matrix and a point target matrix containing equal amounts of point cloud data, and determining the translation matrix T and rotation matrix R required to convert the point source matrix B to the point target matrix A through a coarse alignment step using the singular value decomposition method and a fine alignment step using the ICP algorithm or the Gauss-Newton method. In this way, the advantages of the singular value decomposition method in the coarse alignment stage, which is fast, efficient and highly accurate, can be fully utilized, the number of iterations of the ICP algorithm or the Gauss-Newton method can be effectively reduced, and the convergence speed of the solution process can be accelerated, thereby quickly, efficiently and accurately realizing the spatial calibration of millimeter-wave radars and lidars.
[0011] According to an embodiment of the present application, the obtained translation matrix T and rotation matrix R are used as initial variables, and the translation matrix T and the rotation matrix R can be iteratively updated through a point-to-point iterative nearest point algorithm.
[0012] According to another embodiment of the present application, step S3 may include:
[0013] - Step S31: using the obtained translation matrix T and rotation matrix R as initial variables, obtaining the value of the error function E constructed based on the point cloud data of the point source matrix B, the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R;
[0014] and
[0015] -Step S32: Based on the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R are updated by the Zylesky decomposition method, and the process returns to step S31 to obtain the value of the error function E until the value of the error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
[0016] According to another embodiment of the present application, the obtained translation matrix T and rotation matrix R are used as initial variables, and the translation matrix T and the rotation matrix R can be iteratively updated by the Gauss-Newton method.
[0017] According to another embodiment of the present application, step S3 may include:
[0018] - Step S33: using the obtained translation matrix T and rotation matrix R as initial variables, obtaining the value of the error function E constructed based on the point cloud data of the point source matrix B, the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R;
[0019] -Step S34: Based on the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R are updated by the Gauss-Newton method, and the process returns to step S33 to obtain the value of the error function E until the value of the error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
[0020] According to another embodiment of the present application, the first radar sensor may be configured as a millimeter-wave radar, wherein the point cloud data with the highest echo signal strength in the first point cloud data about each calibration object collected by the millimeter-wave radar are selected as the point cloud data of the point source matrix B assigned to the corresponding calibration object; the second radar sensor may be configured as a laser radar, wherein the geometric center of each calibration object is respectively calculated based on the second point cloud data about each calibration object collected by the laser radar, and the geometric center is respectively used as the point cloud data of the point target matrix A assigned to the corresponding calibration object.
[0021] According to another embodiment of the present application, the first radar sensor may be configured as a laser radar, wherein the geometric center of each calibration object is respectively calculated based on the first point cloud data about each calibration object collected by the laser radar, and the geometric center is respectively used as the point cloud data of the point source matrix B assigned to the corresponding calibration object; the second radar sensor may be configured as a millimeter wave radar, wherein the point cloud data with the highest echo signal intensity in the second point cloud data about each calibration object collected by the millimeter wave radar is selected as the point cloud data of the point target matrix A assigned to the corresponding calibration object.
[0022] According to another embodiment of the present application, the calibration object can be configured as a corner reflector, and a plate with a predetermined reflectivity is constructed on the surface of the corner reflector, wherein the geometric center of the plate is aligned with the geometric center of the surface of the corner reflector, wherein the plate can be configured, for example, as a triangle, a square or a polygon, etc.
[0023] According to a second aspect of the present application, a system for spatial calibration of a radar sensor is provided, the system comprising the following components:
[0024] - a first radar sensor configured to collect first point cloud data about at least three calibration objects;
[0025] - a second radar sensor configured to collect second point cloud data about at least three calibration objects; and
[0026] - A control unit configured to perform the method according to the application.
[0027] According to another embodiment of the present application, the first radar sensor may be configured as a millimeter wave radar, and the second radar sensor may be configured as a laser radar.
[0028] According to another embodiment of the present application, the first radar sensor may be configured as a laser radar, and the second radar sensor may be configured as a millimeter wave radar.
[0029] According to a third aspect of the present application, a vehicle is provided, comprising a system according to the present application.
[0030] According to a fourth aspect of the present application, a computer program product, such as a computer-readable program carrier, is provided, which contains or stores computer program instructions, and when the computer program instructions are executed by a processor, they at least assist in implementing the steps of the method described in the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The principles, features and advantages of the present application may be better understood by describing the present application in more detail below with reference to the accompanying drawings. The accompanying drawings show:
[0032] Figure 1 A flowchart showing a method for spatial calibration of a radar sensor according to an exemplary embodiment of the present application is shown;
[0033] Figure 2 A schematic diagram showing a propagation path of a radar wave on the surface of a corner reflector according to an exemplary embodiment of the present application;
[0034] Figure 3 A flowchart showing a method for spatial calibration of a radar sensor according to another exemplary embodiment of the present application is shown;
[0035] Figure 4 A flowchart showing a method for spatial calibration of a radar sensor according to another exemplary embodiment of the present application; and
[0036] Figure 5 A schematic block diagram of a system for spatial calibration of a radar sensor according to an exemplary embodiment of the present application is shown. DETAILED DESCRIPTION
[0037] In order to make the technical problems, technical solutions and beneficial technical effects to be solved by the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and multiple exemplary embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the scope of protection of the present application.
[0038] Figure 1The flowchart of the method for spatial calibration of a radar sensor according to an exemplary embodiment of the present application is shown. The following exemplary embodiments describe the method according to the present application in more detail.
[0039] The method may include steps S1 to S3. In step S1, a point source matrix B is constructed based on first point cloud data about at least three calibration objects collected by the first radar sensor 11, and a point target matrix A is constructed based on second point cloud data about the at least three calibration objects collected by the second radar sensor 12, wherein the point source matrix B and the point target matrix A contain an equal amount of point cloud data.
[0040] The method according to the present application is performed in a constructed test scene, and at least three calibration objects are arranged at different positions of the test scene, and the at least three calibration objects are located in the common detection area of the first radar sensor and the second radar sensor. The calibration object is particularly configured as a corner reflector, and a plate with a predetermined reflectivity is constructed on the surface of the corner reflector, wherein the geometric center of the plate is aligned with the geometric center of the surface of the corner reflector. Exemplarily, the plate can be configured in a shape such as a triangle, a square or a polygon. The radar wave sent by the radar sensor will be reflected at the end of the corner reflector after scanning the corner reflector, thereby generating a strong echo signal, which can be detected by various types of radar sensors such as millimeter wave radar, laser radar, ultrasonic radar, etc.
[0041] For example, the first radar sensor 11 may be configured as a millimeter wave radar. Figure 2 The schematic diagram of the propagation path of radar waves on the surface of an exemplary corner reflector 21 is shown. The propagation path (indicated by a dotted arrow) of the radar echo reflected back to the millimeter-wave radar at the top O of the corner reflector 21 is the shortest, and the signal strength of the radar echo is also the highest. Other millimeter waves emitted by the millimeter-wave radar will be reflected multiple times on the surface of the corner reflector 21 and reflected back to the millimeter-wave radar along an outgoing path that is antiparallel to the incident path, which causes the propagation paths of these radar echoes (indicated by a solid arrow) to be relatively long, and then causes the signal strength of these radar echoes to be attenuated to a certain extent. Here, the point cloud data with the highest echo signal strength in the first point cloud data about each calibration object collected by the millimeter-wave radar can be selected as the point cloud data of the point source matrix B assigned to the corresponding calibration object, that is, the point cloud data contained in the point source matrix B respectively corresponds to the point cloud data of the radar echo reflected at the top O of each corner reflector.
[0042] Exemplarily, the second radar sensor 12 can be configured as a laser radar. Due to the accuracy and density of the point cloud of the laser radar, a large amount of point cloud data reflected at various surface parts of the corner reflector 21 can be collected by the laser radar. Here, based on the point cloud data of each calibration object collected by the laser radar, the geometric center of each calibration object is respectively obtained, and the geometric center is used as the point cloud data of the point target matrix A assigned to the corresponding calibration object, so that each calibration object corresponds to a point cloud data of the point source matrix B and a point cloud data of the point target matrix A, thereby ensuring that the amount of point cloud data contained in the point source matrix B and the point target matrix A always remains equal.
[0043] Optionally, the first radar sensor 11 can also be configured as a laser radar, and the geometric center of each calibration object is respectively calculated based on the first point cloud data about each calibration object collected by the laser radar, and the geometric center is respectively used as the point cloud data of the point source matrix B assigned to the corresponding calibration object; the second radar sensor 12 can also be configured as a millimeter wave radar, and the point cloud data with the highest echo signal intensity in the second point cloud data about each calibration object collected by the millimeter wave radar is selected as the point cloud data of the point target matrix A assigned to the corresponding calibration object.
[0044] The point cloud data acquired by the radar sensor usually has six degrees of freedom, namely, three physical quantities for characterizing spatial translation (in terms of spatial coordinates t x ,t y ,t z ), and three physical quantities used to characterize spatial flipping, for example: a horizontal angle α, which represents the horizontal angle of the line between the radar sensor and the calibration object. The detection field of view of the millimeter-wave radar is usually within a horizontal angle range of ±60°; a pitch angle β, which represents the vertical angle of the line between the radar sensor and the calibration object, which can vary in the range of -90° to 90°; and a flip angle γ, which cannot be obtained through detection by the radar sensor. Considering that these radar sensors are all fixed by mounting brackets, the roll angle is fixed and can be optionally defaulted to zero.
[0045] In the spatial calibration process of the first radar sensor 11 and the second radar sensor 12, the point source matrix B can be converted into the point target matrix A by rotating and translating the point source matrix B. This process can be expressed by the following formula 1:
[0046] A=B*R+T (Formula 1)
[0047] The rotation matrix T used to characterize the rotation process of the point source matrix B can be determined by the horizontal angle α, pitch angle β and flip angle γ of the radar sensor, and the translation matrix T used to characterize the translation process of the point source matrix B can be determined by the lateral translation amount t x , longitudinal translation t y and height translation t z Next, the rotation matrix R and the translation matrix T are determined through the coarse registration step S2 and the fine registration step S3, thereby completing the spatial calibration of the millimeter wave radar and the laser radar.
[0048] In step S2, the translation matrix T and the rotation matrix R required to transform the point source matrix B into the point target matrix A are obtained by singular value decomposition (SVD). Here, a symmetric matrix H is constructed based on the point target matrix A and the point source matrix B, and the symmetric matrix H is decomposed into the product sum of two orthogonal bases U and V by the following formula 2:
[0049] H=U·∑·V T (Formula 2)
[0050] Among them, V T represents the transpose of the matrix V, and ∑ represents the cumulative summation operator of the matrix. Through the symmetric matrix H and the transpose of the symmetric matrix H T The product of can construct a matrix related only to the orthogonal basis U or a matrix related only to the orthogonal basis V, respectively. This process can be expressed by formula 3 and formula 4:
[0051] H.H T =U·∑·V T ·V·∑ T ·U T =U·∑·∑ T ·U T (Formula 3)
[0052] H T ·H=V·∑ T ·U T ·U·∑·V T =V·∑ T ∑ V T (Formula 4)
[0053] By using the self-adjoint eigenvalue method for formula 3 and formula 4, the eigenvalue u associated with the orthogonal basis U can be obtained. i and the eigenvector v associated with the orthogonal basis V i , where all eigenvectors u iAn orthogonal basis U can be formed, and all eigenvectors vi can form an orthogonal basis V. The rotation matrix R can be obtained by using the obtained orthogonal basis U and V through the Cauchy-Schwarz inequality theorem. The process can be expressed by the following formula 5:
[0054] R=V·U T (Formula 5)
[0055] The translation matrix T can be obtained by formula 1 through the obtained rotation matrix R and the known point source matrix B and point target matrix A. Compared with the process of rough registration by manual operation, the rotation matrix R and translation matrix T required for rough registration of the point source matrix B to the point target matrix A can be determined more efficiently, in a relatively short time and with relatively high accuracy through the singular value decomposition method, thus laying the foundation for the rapid convergence of the fine registration process.
[0056] In step S3, the obtained translation matrix T and rotation matrix R are used as initial variables, and the translation matrix T and the rotation matrix R can be iteratively updated. After the coarse alignment is achieved by the singular value decomposition method, the translation matrix T and the rotation matrix R can be iteratively updated by the point-to-point iterative closest point (ICP) algorithm. Specifically, the obtained translation matrix T and rotation matrix R can be used as initial variables for the first iteration of the ICP algorithm, and the degree of freedom matrix S of the point cloud data can be further precisely aligned through the iterative process of the ICP algorithm, wherein the degree of freedom matrix S is used to characterize the three physical quantities t of the spatial translation. x ,t y ,t z , as well as three physical quantities used to characterize spatial flipping: horizontal angle α, pitch angle β and flip angle γ.
[0057] The following combination Figure 3 The illustrated workflow diagram of a method for spatial calibration of a radar sensor according to another exemplary embodiment of the present application explains in detail the fine registration process of the degree of freedom matrix S using the ICP algorithm.
[0058] like Figure 3 As shown, the step S3 may include steps S31 and S32. In step S31, the obtained translation matrix T and rotation matrix R are used as initial variables to obtain the value of the error function E constructed based on the point cloud data of the point source matrix B, the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R. The error function E can be expressed by the following formula 6:
[0059]
[0060] Among them, bi Represents the point cloud data of the point source matrix B, a i Represents the point cloud data of the point target matrix A.
[0061] In step S32, based on the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R are updated, for example, by the Zyleski decomposition method, and the process returns to step S31 to obtain the value of the error function E until the value of the error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
[0062] In order to simplify the calculation, it can be assumed that the horizontal angle α, the pitch angle β and the flip angle γ are all close to zero, so the correction rotation matrix in the ICP algorithm can be and the corrected translation matrix Simplified to the following formulas 7 and 8 respectively:
[0063]
[0064]
[0065] According to formula 1, the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, the corrected rotation matrix and the corrected translation matrix The mathematical relationship between them can be expressed as the following formula 9:
[0066]
[0067] The three physical quantities t contained in the degree of freedom matrix S used to characterize the spatial translation x ,t y ,t z , and the three physical quantities used to characterize the spatial flip, the horizontal angle α, the pitch angle β and the flip angle γ, are substituted into Formula 9, and the following Formula 10 can be obtained through matrix operation:
[0068]
[0069] Formula 10, which is a linear equation, can be solved by the Zyleski decomposition method, and the solution can be expressed by the following formula 11:
[0070] x=(M T M)-1·M T N(Formula 11)
[0071] Where M represents the matrix in formula 10 M T represents the transpose of the matrix M, and x represents the matrix in formula 10 N represents the matrix in formula 10
[0072] In this way, the three physical quantities t contained in the matrix x that are used to characterize the spatial translation can be x ,t y ,t z The three physical quantities horizontal angle α, pitch angle β and flip angle γ used to characterize spatial flipping are precisely aligned, and the translation matrix T and the rotation matrix R are updated using the obtained x, that is, the original translation matrix T and the rotation matrix R are replaced by the obtained translation matrix T and the rotation matrix R, and then return to step S31 to obtain the value of the error function E until the value of the obtained error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
[0073] Optionally, the obtained translation matrix T and rotation matrix R are used as initial variables, and the translation matrix T and the rotation matrix R can also be iteratively updated by the Gauss-Newton method. Figure 4 A flowchart showing a method for spatial calibration of a radar sensor according to another exemplary embodiment of the present application elaborates on the fine registration process of the degree of freedom matrix S using the Gauss-Newton method.
[0074] like Figure 4 As shown, the step S3 may include steps S33 and S34. In step S33, the obtained translation matrix T and rotation matrix R are used as initial variables to obtain the value of the error function E constructed based on the point cloud data of the point source matrix B, the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R. The error function E can also be expressed by the above formula 6.
[0075] In step S34, the translation matrix T and the rotation matrix R are updated by the Gauss-Newton method based on the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, and the process returns to step S33 to obtain the value of the error function E until the value of the error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
[0076] To illustrate the iterative process of the Gauss-Newton method, it can be assumed that the error function E is expressed as the following formula 12:
[0077]
[0078] Among them, r(R,T) represents the monomial in the error function E.
[0079] The translation matrix T and the rotation matrix R can be iteratively solved according to the following Newton iteration formula about the error function E:
[0080]
[0081] Where H represents the Hessian matrix of r(R,T), represents the gradient of r(R,T), E k represents the error function value of the kth iteration, E k+1 The iterative process may be performed based on the Newton iteration formula until the value of the error function E obtained drops to a predetermined threshold value or reaches a predetermined number of iterations.
[0082] According to an embodiment of the present application, a point source matrix and a point target matrix containing equal amounts of point cloud data are constructed, and the translation matrix T and rotation matrix R required to convert the point source matrix B to the point target matrix A are determined by using a coarse alignment step using a singular value decomposition method and a fine alignment step using an ICP algorithm or a Gauss-Newton method. This allows full use of the advantages of the singular value decomposition method in the coarse alignment stage, which is fast, efficient and highly accurate, to effectively reduce the number of iterations of the ICP algorithm or the Gauss-Newton method, and accelerate the convergence speed of the solution process, thereby quickly, efficiently and accurately achieving the spatial calibration of millimeter-wave radars and lidars.
[0083] In addition, it should be noted that the step numbers described herein do not necessarily represent a chronological order, but are merely a figure mark. The order can be changed according to specific circumstances as long as the technical purpose of the present application can be achieved.
[0084] Figure 5 A schematic block diagram of a system 1 for spatial calibration of a radar sensor according to an exemplary embodiment of the present application is shown.
[0085] like Figure 5 As shown, the system 1 may include the following components:
[0086] - a first radar sensor 11, which is configured to collect first point cloud data about at least three calibration objects;
[0087] - a second radar sensor 12 configured to collect second point cloud data about at least three calibration objects; and
[0088] - A control unit 13 configured to carry out the method according to the present application.
[0089] Optionally, the first radar sensor 11 may be configured as a millimeter wave radar, and the second radar sensor 12 may be configured as a laser radar.
[0090] Optionally, the first radar sensor 11 may be configured as a laser radar, and the second radar sensor 12 may be configured as a millimeter wave radar.
[0091] According to another embodiment of the present application, a vehicle is further provided, wherein the vehicle includes the system 1 .
[0092] If an embodiment includes an “and / or” relationship between a first feature and a second feature, it should be interpreted as follows: according to one embodiment, the embodiment has both the first feature and the second feature, and according to another embodiment, the embodiment has either only the first feature or only the second feature.
[0093] Although specific embodiments have been described above, these embodiments are not intended to limit the scope of the present disclosure, even when only a single embodiment is described with respect to specific features. The feature examples provided in the present disclosure are intended to be illustrative, not restrictive, unless otherwise stated. In specific implementations, multiple features may be combined with each other, where technically feasible, according to actual needs. Various substitutions, changes, and modifications may also be conceived without departing from the spirit and scope of the present application.
Claims
1. A method for spatial calibration of a radar sensor, the method comprising: Step S1: constructing a point source matrix B based on first point cloud data about at least three calibration objects collected by a first radar sensor (11), and constructing a point target matrix A based on second point cloud data about the at least three calibration objects collected by a second radar sensor (12), wherein the point source matrix B and the point target matrix A contain an equal amount of point cloud data; Step S2: Obtaining the translation matrix T and rotation matrix R required to transform the point source matrix B into the point target matrix A by singular value decomposition method; and Step S3: Using the obtained translation matrix T and rotation matrix R as initial variables, iteratively update the translation matrix T and the rotation matrix R.
2. The method according to claim 1, wherein: The obtained translation matrix T and rotation matrix R are used as initial variables, and the translation matrix T and the rotation matrix R are iteratively updated through a point-to-point iterative nearest point algorithm.
3. The method according to claim 2, wherein: The step S3 comprises: Step S31: taking the obtained translation matrix T and rotation matrix R as initial variables, obtaining the value of the error function E constructed based on the point cloud data of the point source matrix B, the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R; and Step S32: Based on the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R are updated by the Zylesky decomposition method, and the process returns to step S31 to obtain the value of the error function E until the value of the error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
4. The method according to claim 1, wherein: The obtained translation matrix T and rotation matrix R are used as initial variables, and the translation matrix T and the rotation matrix R are iteratively updated by the Gauss-Newton method.
5. The method according to claim 4, wherein: The step S3 comprises: Step S33: using the obtained translation matrix T and rotation matrix R as initial variables, obtaining a value of an error function E constructed based on the point cloud data of the point source matrix B, the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R; and Step S34: Based on the point cloud data of the point source matrix B and the point cloud data of the point target matrix A, the translation matrix T and the rotation matrix R are updated by the Gauss-Newton method, and the process returns to step S33 to obtain the value of the error function E until the value of the error function E drops to a predetermined threshold or reaches a predetermined number of iterations.
6. The method according to any one of claims 1 to 5, wherein: The first radar sensor (11) is configured as a millimeter-wave radar, wherein point cloud data with the highest echo signal strength in the first point cloud data about each calibration object collected by the millimeter-wave radar are selected as the point cloud data of the point source matrix B assigned to the corresponding calibration object, and the second radar sensor (12) is configured as a laser radar, wherein the geometric center of each calibration object is respectively calculated based on the second point cloud data about each calibration object collected by the laser radar, and the geometric center is respectively used as the point cloud data of the point target matrix A assigned to the corresponding calibration object.
7. The method according to any one of claims 1 to 5, wherein: The first radar sensor (11) is configured as a laser radar, wherein the geometric center of each calibration object is respectively calculated based on the first point cloud data about each calibration object collected by the laser radar, and the geometric center is respectively used as the point cloud data of the point source matrix B assigned to the corresponding calibration object, and the second radar sensor (12) is configured as a millimeter wave radar, wherein the point cloud data with the highest echo signal intensity in the second point cloud data about each calibration object collected by the millimeter wave radar is selected as the point cloud data of the point target matrix A assigned to the corresponding calibration object.
8. The method according to any one of claims 1 to 5, wherein: The calibration object is configured as a corner reflector, and a plate with a predetermined reflectivity is constructed on the surface of the corner reflector, wherein the geometric center of the plate is aligned with the geometric center of the surface of the corner reflector, and wherein the plate is configured as a triangle, a square, or a polygon.
9. A system (1) for spatial calibration of a radar sensor, the system (1) comprising the following components: A first radar sensor (11) configured to collect first point cloud data about at least three calibration objects; A second radar sensor (12) configured to collect second point cloud data about at least three calibration objects; and A control unit (13) configured to execute the method according to any one of claims 1 to 8.
10. The system (1) according to claim 9, wherein: The first radar sensor (11) is configured as a millimeter wave radar, and the second radar sensor (12) is configured as a laser radar.
11. The system (1) according to claim 9, wherein: The first radar sensor (11) is configured as a laser radar, and the second radar sensor (12) is configured as a millimeter wave radar.
12. A vehicle comprising a system (1) according to any one of claims 9 to 11. 13 . A computer program product, such as a computer-readable program carrier, comprising or storing computer program instructions, which, when executed by a processor, at least assist in implementing the steps of the method according to claim 1 .
