A method and system for SVD multi-mechanical laser radar extrinsic calibration based on the same field of view

Abstract

The present invention uses a self-built coordinate system and plane fitting to project the identified radar corner points to the corner points of the real object under the self-built coordinates, thereby compensating for the missing part of the point cloud due to laser density and establishing a relatively complete implicit relationship. If the fields of view of different laser radars are small, external parameter calibration can be performed by moving the same marker in the fields of view of the two laser radars, and more sets of data can be collected to achieve a more accurate calibration result.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method and system for calibrating external parameters of an SVD multi-mechanical laser radar based on the same field of view. Background Art

[0002] Among the existing mechanical multi-lidar external parameter calibration methods, the NDT algorithm is the most widely used and is widely used in mechanical multi-lidar external parameter calibration. However, the NDT algorithm has the problem that when the initial error of the matching point cloud is too large or the initial angle difference is large, the registration is prone to fall into the local optimal solution, resulting in large errors in the posture and position calculated by the registration. This will make the fusion effect of the lidar image at a distance poor, and it is difficult to expand the multi-sensor perception range of autonomous driving, which poses a great hidden danger to the safety of autonomous driving. Existing improvement methods mostly use the method of increasing feature points to optimize the situation where the NDT algorithm cannot obtain the optimal solution during iteration and cannot be calibrated quickly, which makes it difficult to adapt to the needs of large-scale calibration in engineering. The shortcomings of common technical solutions in the prior art are as follows:

[0003] 1) Using the ndt algorithm, the calculation time is long because it is impossible to obtain a more accurate initial value, the data is large and it takes a long time to obtain the optimal solution.

[0004] 2) When collecting point cloud images at different coordinates, due to the different angles and densities of the laser, it is easy to have problems with missing or mismatched feature points of the point cloud images of the same object.

[0005] 3) The ndt algorithm is suitable for multi-lidar calibration under the same field of view. If the two point cloud images have little overlap, it is impossible to calculate more accurate external parameter data. Summary of the invention

[0006] The present invention calibrates mechanical multi-lidars under the same field of view based on SVD, avoiding the disadvantage that the traditional ndt algorithm falls into the process of local optimal solution, resulting in a long calculation time.

[0007] Therefore, the technical solution provided by the present invention is as follows:

[0008] As a first aspect, the present invention provides an SVD multi-mechanical laser radar extrinsic parameter calibration method based on the same field of view, the method steps are as follows:

[0009] S1: Use LiDAR A and LiDAR B to simultaneously collect point cloud data of the black and white grid calibration plate in the same field of view.

[0010] S2, extract the four boundary corner points of the calibration plate from the collected point cloud data, and obtain the boundary four corner point coordinates 1 of LiDAR A and the boundary four corner point coordinates 2 of LiDAR B. Measure the length and width of the calibration plate, establish a self-built coordinate system with its center point as the origin of the coordinate system, and obtain the boundary four corner point coordinates 3 of the calibration plate.

[0011] S3, obtain the coordinates of the four corner points of the boundary one, the four corner points of the boundary two, and the four corner points of the boundary three, use SVD to make a one-to-one correspondence between the data and perform singular value decomposition, obtain the rotation matrix of the corresponding coordinates, and obtain the external reference function between laser radar A and laser radar B through the rotation matrix.

[0012] S4, the target point cloud is rigidly transformed through the rotation matrix to complete the external parameter calibration.

[0013] In combination with the first aspect, in a first case of any possible case, step S1 is:

[0014] On the premise of ensuring that the calibration plate is presented vertically and completely in the point cloud image, laser radar A and laser radar B are used to simultaneously collect point cloud data of the black and white grid calibration plate in the same field of view.

[0015] In combination with the first aspect or the first situation described above, in any of the possible situations, the second situation is that the method of extracting the four boundary corner points of the calibration plate from the collected point cloud data in S2 to obtain the boundary four corner point coordinates 1 of the laser radar A and the boundary four corner point coordinates 2 of the laser radar B is:

[0016] First, the Cropbox function is used to cut the calibration plate into cubes, and the point cloud of the cube is plane-segmented using random sampling consistency.

[0017] After obtaining the calibration plate point cloud plane, obtain the neighboring points of the sampling points as local surface reference data, fit the micro-cutting plane of the data by the least squares method, and project it onto the micro-cutting plane;

[0018] Identify the boundary features of scattered point clouds based on the maximum angle between the sampling point and its k nearest neighbors’ corresponding projection points;

[0019] Then, the coordinates of the four corner points of the plane are obtained by sorting the coordinates, so that the coordinates of the four corner points of the boundary of laser radar A and the coordinates of the four corner points of the boundary of laser radar B are obtained.

[0020] In combination with the first aspect or the first and second situations, in a third situation in any possible situation, step S3 is:

[0021] Get the coordinates 1, 2 and 3 of the four corner points of the boundary, use SVD to make a one-to-one correspondence between the data and perform singular value decomposition, get the rotation matrix of the corresponding coordinates, and use the rotation matrix to obtain the external parameter function between laser radar A and laser radar B. The specific implementation process is: first, perform SVD decomposition on the coordinates 1 of the four corner points of the boundary of laser radar A and the coordinates 3 of the four corner points of the boundary of the calibration plate in the self-built coordinate system to obtain the rotation matrix M1, and then perform SVD decomposition on the coordinates 3 of the four corner points of the boundary of the calibration plate in the self-built coordinate system and the coordinates 2 of the four corner points of the boundary of laser radar B to obtain the rotation matrix M2, and obtain the external parameter function between laser radar A and laser radar B as: M=M1*M2.

[0022] As a second aspect, the present invention provides an SVD multi-mechanical laser radar extrinsic calibration system based on the same field of view, the system comprising:

[0023] The laser radar group consists of multiple laser radars, which use multiple laser radars to collect point cloud data of the black and white grid calibration plate in the same field of view;

[0024] A memory for storing programs that can be run by the processor;

[0025] A processor is used to run the program stored in the memory to implement any of the above-mentioned methods for calibrating extrinsic parameters of SVD multi-mechanical laser radars based on the same field of view.

[0026] As a third aspect, the present invention provides an SVD multi-mechanical laser radar external parameter calibration device based on the same field of view, the device comprising: a laser radar device, a computer device, a wired or wireless data communication connection is established between the laser radar device and the computer device, and the computer device runs a program to implement any of the above-mentioned SVD multi-mechanical laser radar external parameter calibration methods based on the same field of view.

[0027] As a fourth aspect, the present invention provides an autonomous driving vehicle, which is equipped with the above-mentioned SVD multi-mechanical laser radar external parameter calibration device based on the same field of view.

[0028] The beneficial effects of the present invention are:

[0029] By selecting self-built coordinate system and plane fitting, the problem of LiDAR’s difficulty in accurately selecting corner points due to the density of laser scanning surface is solved. At the same time, the target point cloud is filtered, segmented, and down-sampled, and the region of interest is reasonably set to reduce the amount of calculation. The calculation speed and calibration accuracy are greatly improved by solving the rotation matrix through SVD decomposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 A flow chart of the calibration scheme of the present invention;

[0032] Figure 2 This is a rendering of the calibration solution of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention.

[0034] The technical solution of the present invention is described below in conjunction with the accompanying drawings and embodiments. The meanings of some English abbreviations are as follows:

[0035] SVD--Singular Value Decomposition

[0036] RSC--Random Sample Consensus

[0037] OLS--Ordinary Least Squares

[0038] Embodiment 1:

[0039] like Figure 1 As shown, this embodiment provides an SVD multi-mechanical laser radar extrinsic parameter calibration method based on the same field of view, and the method steps are as follows:

[0040] S1: On the premise of ensuring that the calibration plate is vertically and completely presented in the point cloud image, use LiDAR A and LiDAR B to simultaneously collect point cloud data of the black and white grid calibration plate in the same field of view.

[0041] S2, extract the four boundary corner points of the calibration plate from the collected point cloud data, and obtain the boundary four corner point coordinates 1 of laser radar A and the boundary four corner point coordinates 2 of laser radar B. First, use the Cropbox function to cut the calibration plate range into a cube, and use the random sampling consistency to plane segment the point cloud of the cube. After obtaining the calibration plate point cloud plane, obtain the neighboring points of the sampling point as the local surface reference data, fit the micro-cut plane of the data by the least squares method, and project it onto the micro-cut plane. Identify the boundary features of the scattered point cloud according to the maximum angle between the projection points corresponding to the sampling point and its k nearest neighbors. Then obtain the coordinates of the four corner points of the plane by sorting the coordinates, so as to obtain the boundary four corner point coordinates 1 of laser radar A and the boundary four corner point coordinates 2 of laser radar B. Measure the length and width of the calibration plate, establish a self-built coordinate system with its center point as the origin of the coordinate system, and obtain the boundary four corner point coordinates 3 of the calibration plate.

[0042] S3, obtain the coordinates 1, 2 and 3 of the four corner points of the boundary, use SVD to make a one-to-one correspondence between the data and perform singular value decomposition, obtain the rotation matrix of the corresponding coordinates, and obtain the external parameter function between laser radar A and laser radar B through the rotation matrix. The specific implementation process is: first, perform SVD decomposition on the coordinates 1 of the four corner points of the boundary of laser radar A and the coordinates 3 of the four corner points of the boundary of the calibration plate in the self-built coordinate system to obtain the rotation matrix M1, and then perform SVD decomposition on the coordinates 3 of the four corner points of the boundary of the calibration plate in the self-built coordinate system and the coordinates 2 of the four corner points of the boundary of laser radar B to obtain the rotation matrix M2. The external parameter function between laser radar A and laser radar B is obtained as: M=M1*M2.

[0043] S4, the target point cloud is rigidly transformed through the rotation matrix to complete the external parameter calibration.

[0044] Different from the disadvantage of the NDT algorithm that it takes a long time to calculate, the present invention calculates the rotation matrix by extracting the corner points of the same object from more than three different point clouds, which has a small amount of calculation and can quickly calculate the external parameter data. By building a coordinate system and plane fitting, the identified radar corner points are projected to the corner points of the real object under the self-built coordinates, which compensates for the missing part of the point cloud due to laser density and establishes a more complete implicit relationship. If the field of view of different laser radars is small, the external parameter calibration can be performed by moving the same marker in the field of view of the two laser radars, and more sets of data can be collected to achieve a more accurate calibration result.

[0045] like Figure 2 As shown, the dark color is the point cloud calibrated by laser radar A, and the light color is the point cloud calibrated by laser radar B. The two are fused together in a relatively consistent manner. The external parameters of the two radars can be obtained by running the program tzz function package in 50ms. Compared with the ndt calibration algorithm which takes 3s, the present invention is more efficient.

[0046] Example 2

[0047] This embodiment provides an SVD multi-mechanical laser radar extrinsic parameter calibration system based on the same field of view, the system comprising:

[0048] The laser radar group consists of multiple laser radars, which use multiple laser radars to collect point cloud data of the black and white grid calibration plate in the same field of view;

[0049] A memory for storing programs that can be run by the processor;

[0050] A processor is used to run the program stored in the memory to implement any of the above-mentioned methods for calibrating extrinsic parameters of SVD multi-mechanical laser radars based on the same field of view.

[0051] Example 3

[0052] This embodiment provides an SVD multi-mechanical laser radar external parameter calibration device based on the same field of view, the device includes: a laser radar device, a computer device, a wired or wireless data communication connection is established between the laser radar device and the computer device, and the computer device runs a program to implement any of the above-mentioned SVD multi-mechanical laser radar external parameter calibration methods based on the same field of view.

[0053] Example 4

[0054] This embodiment provides an autonomous driving vehicle, which is equipped with the above-mentioned SVD multi-mechanical laser radar external parameter calibration device based on the same field of view.

[0055] It should be understood that the above embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content of the present invention, those skilled in the art can make various changes and modifications to the present invention, and these equivalent forms also fall within the scope limited by the appended claims of the present application.

Claims

1. A method for calibrating extrinsic parameters of SVD multi-mechanical laser radars based on the same field of view, characterized in that: The method steps are as follows: S1. Use LiDAR A and LiDAR B to simultaneously collect point cloud data of the black and white grid calibration plate in the same field of view; S2. Extract the four boundary corner points of the calibration plate from the collected point cloud data, and obtain the coordinates 1 of the four boundary corner points of the laser radar A and the coordinates 2 of the four boundary corner points of the laser radar B; measure the length and width of the calibration plate, establish a self-built coordinate system with its center point as the origin of the coordinate system, and obtain the coordinates 3 of the four boundary corner points of the calibration plate; S3. Obtain the coordinates of the four corner points of the boundary, the coordinates of the four corner points of the boundary, and the coordinates of the four corner points of the boundary, use SVD to make a one-to-one correspondence between the data and perform singular value decomposition, obtain the rotation matrix of the corresponding coordinates, and obtain the external parameter function between the laser radar A and the laser radar B through the rotation matrix; S4. Perform rigid body transformation on the target point cloud through the rotation matrix to complete the external parameter calibration; The method of extracting the four boundary corner points of the calibration plate from the collected point cloud data in S2 to obtain the boundary four corner point coordinates 1 of the laser radar A and the boundary four corner point coordinates 2 of the laser radar B is: First, the Cropbox function is used to cut the calibration plate into cubes, and the point cloud of the cube is plane-segmented using random sampling consistency. After obtaining the calibration plate point cloud plane, obtain the neighboring points of the sampling points as local surface reference data, fit the micro-cutting plane of the data by the least squares method, and project it onto the micro-cutting plane; Identify the boundary features of scattered point clouds based on the maximum angle between the sampling point and its k nearest neighbors’ corresponding projection points; Then, the coordinates of the four corner points of the plane are obtained by sorting the coordinates, so that the coordinates of the four corner points of the boundary of the laser radar A and the coordinates of the four corner points of the boundary of the laser radar B are obtained; The S3 is: Get the coordinates 1, 2 and 3 of the four corner points of the boundary. First, perform SVD decomposition on the coordinates 1 of the four corner points of the boundary of the laser radar A and the coordinates 3 of the four corner points of the boundary of the calibration plate in the self-built coordinate system to obtain the rotation matrix M1. Secondly, perform SVD decomposition on the coordinates 3 of the four corner points of the boundary of the calibration plate in the self-built coordinate system and the coordinates 2 of the four corner points of the boundary of the laser radar B to obtain the rotation matrix M2. The external parameter function between the laser radar A and the laser radar B is: M=M1*M2; S1 is the point cloud data of the black and white grid calibration plate in the same field of view collected simultaneously by laser radar A and laser radar B, under the premise of ensuring that the calibration plate is vertically and completely presented in the point cloud map.

2. A system for implementing the SVD multi-mechanical laser radar extrinsic parameter calibration method based on the same field of view as claimed in claim 1, characterized in that: The system comprises: The laser radar group consists of multiple laser radars, which use multiple laser radars to collect point cloud data of the black and white grid calibration plate in the same field of view; A memory for storing programs that can be run by the processor; A processor is used to run the program stored in the memory to implement the SVD multi-mechanical laser radar external parameter calibration method based on the same field of view described in claim 1.

3. An SVD multi-mechanical laser radar extrinsic parameter calibration device based on the same field of view, characterized in that: The device includes: a laser radar device and a computer device. A wired or wireless data communication connection is established between the laser radar device and the computer device. The computer device runs a program to implement the SVD multi-mechanical laser radar external parameter calibration method based on the same field of view described in claim 1.

4. An autonomous driving vehicle, characterized in that: The vehicle is equipped with an SVD multi-mechanical laser radar external parameter calibration device based on the same field of view as described in claim 3.

Images