A wheel odometer calibration method and lidar external parameter calibration method

Abstract

The present invention discloses a wheeled odometer calibration method and a laser radar external parameter calibration method. The calibration method calculates a position correction proportional coefficient α through a position calibration method and a posture correction proportional coefficient β through a posture calibration method. The calculated position correction proportional coefficient α and posture correction proportional coefficient β are used to obtain a wheeled odometer A according to the wheel speed and are calibrated and corrected by calculating α and β. The external parameter calibration method obtains the calibrated wheeled odometer A, obtains laser point cloud data at two adjacent moments through a laser radar, matches the two laser point cloud data, and estimates the speed and position of a differential wheeled robot to obtain a laser odometer B. Multiple sets of wheeled odometers A and laser odometers B are collected and jointly calibrated to obtain an optimal rotation matrix and an optimal translation matrix, thereby obtaining the required relationship values ​​to complete the joint calibration. The method of the present invention has low requirements for the calibration environment, can improve accuracy, and effectively simplifies the laser radar calibration process.

Description

Technical Field

[0001] The present invention relates to the technical field of laser radar calibration methods. Background Art

[0002] As an important sensor in the field of robotics, LiDAR plays a crucial role in positioning and navigation. LiDAR can provide rich environmental information, while wheeled odometry can provide real-time and continuous robot pose estimation, complementing each other. However, because there are only theoretical models for the LiDAR's installation pose and the robot's base coordinates, and dynamic factors such as robot chassis machining errors, mechanical tolerances, assembly deviations, reducer accuracy, reducer backlash, mass, joint flexibility, and link flexibility, errors inevitably exist between the theoretical and actual models. These errors are primarily due to: first, the actual LiDAR installation pose cannot accurately align with the theoretical extrinsic parameters; second, the robot's center of rotation (i.e., the robot's base coordinate system) is not necessarily located at the corresponding position in the theoretical model (usually the robot's geometric center). The accuracy of these two parameters directly affects the accuracy of the LiDAR's positioning perception results. Therefore, precise calibration of the extrinsic parameters of both is necessary to achieve multi-sensor data fusion to improve positioning and navigation accuracy.

[0003] Existing solutions to this problem include a lidar extrinsic parameter calibration method, device, and storage medium based on point cloud registration, such as Chinese patent application No. 202311409630.X. This method collects a first point cloud and a second point cloud of the vehicle (collected from the left and right wheels, respectively), uses a density clustering method to obtain the target center coordinates, and performs point cloud registration with the target center of a preset lidar coordinate system to obtain the target transformation parameters, thereby obtaining the lidar extrinsic parameters. Although this method reduces the site and environmental requirements for lidar extrinsic parameter calibration and does not require a four-wheel alignment system or measurement of the three-dimensional coordinates of the target center in the vehicle coordinate system, it still requires specific equipment and environment to collect point cloud data. There is also a vehicle-mounted laser radar external parameter calibration system and method such as Chinese patent application No. 202410519238.9, which includes a vehicle fixture, a first reflector, a second reflector, a reflective column, and a level, etc. The system is provided with a reflector and a reflective column in front of the vehicle-mounted laser radar, and the radar waves emitted by the vehicle-mounted laser radar are reflected by the reflector to achieve external parameter calibration; the system requires equipment such as vehicle fixtures, reflectors, and reflective columns to set up the calibration scene, which also increases the complexity of the system and the requirements for the environment. At the same time, the position and direction of the reflector and reflective column need to be accurately set to ensure accurate calibration results. It can be seen that the existing technical solutions are still highly dependent on equipment and the environment, which is not conducive to simplifying the laser radar calibration process. Summary of the Invention

[0004] The present invention aims to provide a wheel odometer calibration method that can improve the accuracy of the wheel odometer. Another object of the present invention is to provide a lidar extrinsic parameter calibration method that has low requirements for the calibration environment, can improve accuracy, and effectively simplifies the lidar calibration process.

[0005] To achieve the above purpose, the technical solution of the present invention is: a wheel odometer calibration method, which calculates the position correction proportional coefficient by a position calibration method , calculate the attitude correction coefficient through the attitude calibration method The wheel odometer A calculates the robot's running speed and relative displacement position through the wheel speed of the differential wheel robot and corrects the proportional coefficient through the position and attitude correction coefficient Perform calculation calibration correction. The method for calculating and calibrating the wheel odometer A includes: assuming the position state of the differential wheel robot is ,in 、 represents the spatial coordinates of the differential wheeled robot, Represents the angle of the differential wheeled robot. According to the motion model of the differential wheeled robot, its posture recursion equation in discrete time is: hour, ,

[0006] when hour ,in, is the speed of the robot at time t, is the angular velocity of the robot at time t, express arrive The time interval between moments, is the position correction proportional coefficient, is the attitude correction coefficient before calibration .

[0007] The position calibration method is to measure the distance between the differential wheeled robot and the preset target The distance from the starting position to the target The distance deviation between the actual stop position and the preset end point after the end position And the distance reading of wheel odometer A and the distance between the target Deviation , calculate the position correction proportional coefficient , the expression is .

[0008] The posture calibration method is to measure the deviation distance between the axis of the differential wheeled robot and the calibration reference line after the axis of the differential wheeled robot is aligned with the calibration reference line for multiple rotations in place, and divide the deviation difference by the distance from the rotation center of the differential wheeled robot to the preset front mark to obtain the actual posture deviation. , the measured odometer attitude reading and target attitude deviation is ,but: .

[0009] A laser radar extrinsic parameter calibration method is disclosed. A calibrated and corrected wheel odometry A is obtained by using the aforementioned wheel odometry calibration method. Laser point cloud data at two adjacent moments are obtained by using a laser radar of a differential wheeled robot. The laser point cloud data at two adjacent moments are matched to estimate the speed and position of the differential wheeled robot, thereby obtaining a laser odometry B. Multiple sets of wheel odometry A and laser odometry B are collected and jointly calibrated to obtain an optimal rotation matrix and an optimal translation vector, thereby obtaining the required extrinsic parameters to complete the joint calibration.

[0010] The method for matching the laser point cloud data of two adjacent moments is to set corresponding laser point cloud data for the two adjacent frames of laser point cloud data. and ,in , laser point cloud data and The centroid is removed to obtain the corresponding point cloud without the centroid and then perform point cloud registration. The centroid removal method is as follows: the centroids of the two frames of laser point clouds are and , then the corresponding laser point cloud without the centroid is and .

[0011] The laser point cloud registration method is: , the matrix Perform SVD decomposition, that is ,in, is a left singular matrix, is a matrix composed of singular values, is a right singular matrix, then the rotation matrix , translation matrix .

[0012] The joint calibration method is to assume that the pose transformation from the differential wheeled robot base coordinate system to the laser radar coordinate system is the external parameter to be determined. ,exist At time , the wheel odometer reading of the differential wheeled robot is , the laser odometry reading is ,exist At this moment, the robot's wheel odometer reading is , the laser odometry reading is ,

[0013] The wheel odometer and laser odometer at two moments have the following conversion relationship: Combining the above two equations, we can get , solve it using a two-step method: ,in R X For external reference The rotation matrix of and are the rotations of transformation matrix A and transformation matrix B respectively, To seek external reference The translation vector of , ,in 、 They are 、 The corresponding Lie algebra can be transformed into: , collect multiple sets of wheel odometer A and laser odometer data B and solve the optimal rotation matrix through least square fitting, which is recorded as R X * , ,Depend on It can be seen that , multiply both sides by ,in I As the unit matrix, rearrange it to solve the optimal translation vector, which is recorded as * , .

[0014] By adopting the above technical solution, the beneficial effect of the present invention is that the above-mentioned laser radar external parameter calibration method of the present invention does not rely on calibration objects such as calibration plates, does not require additional deployment of calibration scenes and other tedious processes, and has low dependence on equipment and environment, that is, the calibration method of the present invention has low requirements for the calibration environment. By combining the wheel odometer and the laser odometer, a two-step hand-eye calibration is adopted to solve the external parameter calibration problem of the laser radar in the differential wheeled robot base coordinate system. Specifically, the wheel odometer is calculated by the encoder and calibrated by the above-mentioned wheel odometer calibration method, the laser odometer is calculated by point cloud registration, and the two-step hand-eye calibration is used to realize the external parameter calibration of the laser radar and the robot base coordinate system. This effectively reduces the dependence on equipment and environment, and can also improve the accuracy of parameters such as the position and posture of the laser radar in the differential wheeled robot base coordinate system. These parameters also directly affect the accuracy of the laser radar positioning perception results, and play a vital role in the robot's positioning, navigation and obstacle avoidance functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention relates to a principle block diagram of a laser radar external parameter calibration method.

[0016] Figure 2 The present invention relates to a position calibration method for laser radar external parameters.

[0017] Figure 3 The present invention relates to a posture calibration method in a laser radar external parameter calibration method.

[0018] Figure 4 The invention relates to a joint calibration method for laser radar external parameter calibration. DETAILED DESCRIPTION

[0019] In order to further explain the technical solution of the present invention, the present invention is described in detail below through specific embodiments.

[0020] A wheel odometer calibration method of the present invention is disclosed in detail by a laser radar external parameter calibration method disclosed below in this embodiment. The principle of a laser radar external parameter calibration method is as follows: Figure 1 As shown, the running speed and relative displacement position of the differential wheeled robot can be calculated by the wheel speed of the differential wheeled robot, and the obtained wheel odometer is recorded as A. Due to errors in processing, installation, measurement, etc., the calculated speed and position are not so accurate. The present invention calibrates and corrects the wheel odometer A through the following calibration coefficients. Specifically, the position correction proportional coefficient is calculated by the position calibration method. , calculate the attitude correction coefficient through the attitude calibration method , corrected by the position proportional coefficient and attitude correction coefficient A calculation calibration correction is performed, as described in detail below.

[0021] In addition, the laser radar of the differential wheeled robot scans the surrounding environment to obtain the laser data at the current moment. After the differential wheeled robot moves, it scans the surrounding environment again to obtain the laser data at the next moment, that is, the laser point cloud data of two adjacent moments are obtained. Then, the laser point cloud data of the two adjacent moments are matched, and the speed and position of the differential wheeled robot are estimated to obtain the laser odometry recorded as B; multiple sets of wheel odometry A and laser odometry B are collected for joint calibration, such as the following corresponding pairs: A1-B1, A2-B2, A3-B3, A4-B4,... to calculate the relative relationship between the laser radar and the robot base coordinate system, and obtain the optimal rotation matrix and the optimal translation matrix, so as to obtain the required external parameters to complete the joint calibration. Here, it is like solving an equation. For the straight line y=kx+c, given two sets of data (0, 2) and (1,3), we can solve k=1 and c=2 to obtain the equation of the straight line, but the unknowns here are rotation and translation.

[0022] The calibration method of the wheel odometer A is as follows:

[0023] Motion solution, assuming the posture state of the differential wheeled robot is ,in 、 represents the spatial coordinates of the differential wheeled robot, Represents the angle of the differential wheeled robot. According to the motion model of the differential wheeled robot, its posture recursion equation in discrete time is as follows: hour, ,when hour, ,in, is the speed of the robot at time t, is the angular velocity of the robot at time t, express arrive The time interval between moments, is the position correction proportional coefficient, is the attitude correction coefficient before calibration .

[0024] The position calibration method is to prepare measuring tools such as a tape measure and a marker pen, and mark the target distance with a certain distance. Make a "one" mark on the ground to mark the starting point and the end point, and then control the differential wheel robot to run to the end point through the program. Figure 2 As shown, the distance deviation between the measured actual position of the differential wheeled robot and the end point is recorded as , when the differential wheeled robot crosses the end point, it is a positive value, and the distance reading of the wheel odometer A and the target distance are recorded. The deviation is recorded as ,but .

[0025] The posture calibration method aligns the center axis of the differential wheeled robot with the calibration reference line, and then controls the differential wheeled robot to rotate multiple times through a program. In this embodiment, two rotations are used as an example. Figure 3 As shown, the deviation distance between the center axis of the differential wheeled robot and the calibration reference line is measured and divided by the distance from the rotation center of the differential wheeled robot to the front mark (a mark pre-made on the front of the differential wheeled robot), which is the actual posture deviation, recorded as When the differential wheeled robot passes the mark, the value is positive, and the deviation between the odometer attitude reading and the target attitude is recorded as ,but: .

[0026] The laser point cloud data matching method for the two adjacent moments is as follows: The laser odometer is realized by matching the laser point cloud data at adjacent moments to estimate the motion state of the differential wheeled robot. For two adjacent frames of laser point cloud data, there are corresponding laser point cloud data. and ,in , laser point cloud data and The centroid is removed to obtain the corresponding point cloud without the centroid and then the point cloud is registered.

[0027] The centroid removal method is as follows: the centroids of the two frames of laser point clouds are and , then the corresponding laser point cloud without the centroid is and .

[0028] The laser point cloud registration method is: , the matrix Perform SVD decomposition, that is ,in, is a left singular matrix, is a matrix composed of singular values, is a right singular matrix, then the rotation matrix , translation matrix .

[0029] The joint calibration method is as follows: Since the base coordinates of the laser radar and the robot are relatively fixed (rigid connection), the posture relationship between the two is constrained by the external parameters of the laser radar installation posture in the base coordinate system. The posture transformation from the differential wheeled robot base coordinate system to the laser radar coordinate system is the external parameter to be determined, denoted as ,like Figure 4 As shown, in At time , the wheel odometer reading of the differential wheeled robot is , the laser odometry reading is ,exist At this moment, the robot's wheel odometer reading is , the laser odometry reading is , in the figure, B 1,2 express P 1 arrive P 2 The position change of the laser odometry, B 2,3 express P 2 arrive P 3 The position change of the laser odometry, A 1,2 express P 1 arrive P 2 The posture change of the wheel odometer, A 2,3 express P 2 arrive P 3 Pose changes of wheel odometry.

[0030] The wheel odometer and laser odometer at two moments have the following conversion relationship, namely It can be directly obtained by wheel odometer solution, or by laser odometer combined with external reference The calculation is as follows: Combining the above two equations, we can get , solve it using a two-step method: Among them and The rotations of transformation matrix A and transformation matrix B respectively, R X To seek external reference The rotation matrix of To seek external reference The translation vector of , ,in 、 They are 、 The corresponding Lie algebra can be transformed into: Therefore, by collecting multiple sets of wheel odometer and laser odometer data, the optimal rotation matrix can be solved by least square fitting, which is recorded as R X * ,Right now , this form is consistent with the laser point cloud registration optimization method of the laser odometry, and is still solved by SVD, which will not be described here.

[0031] Depend on , it can be seen that , in order to ensure The left side is reversible, multiply both sides by the left ,in I As the unit matrix, rearrange it to solve the optimal translation vector, which is recorded as * ,Right now .

[0032] In summary, the calibration method of the present invention does not rely on calibration objects such as calibration plates, does not require additional deployment of calibration scenes and other tedious processes, has low requirements for the calibration environment, and can effectively simplify the calibration process of the lidar.

[0033] The above embodiments and drawings do not limit the product form and style of the present invention. Any appropriate changes or modifications made by ordinary technicians in the relevant technical field should be deemed to be within the patent scope of the present invention.

Claims

1. A wheel odometer calibration method, characterized in that: Calculate the position correction coefficient using the position calibration method , calculate the attitude correction coefficient through the attitude calibration method The wheel odometer A calculates the robot's running speed and relative displacement position through the wheel speed of the differential wheel robot and corrects the proportional coefficient through the position and attitude correction coefficient Perform calculation calibration correction; the method for calculating the calibration correction of the wheel odometer A includes, assuming the position state of the differential wheeled robot is ,in 、 represents the spatial coordinates of the differential wheeled robot, Represents the angle of the differential wheeled robot. According to the motion model of the differential wheeled robot, its posture recursion equation in discrete time is: hour, ,when hour, ,in, is the speed of the robot at time t, is the angular velocity of the robot at time t, express arrive The time interval between moments, is the position correction proportional coefficient, is the attitude correction coefficient before calibration The position calibration method is to measure the differential wheeled robot from the preset target distance The distance from the starting position to the target The distance deviation between the actual stop position and the preset end point after the end position And the distance reading of wheel odometer A and the distance between the target Deviation , calculate the position correction proportional coefficient , the expression is The posture calibration method is to measure the deviation distance between the axis of the differential wheeled robot and the calibration reference line after the axis of the differential wheeled robot is aligned with the calibration reference line for multiple rotations in place, and divide the deviation distance by the distance from the rotation center of the differential wheeled robot to the preset front mark to obtain the actual posture deviation. , the measured odometer attitude reading and target attitude deviation is ,but: .

2. A laser radar external parameter calibration method, characterized in that: A calibrated and corrected wheel odometry A is obtained by a wheel odometry calibration method as described in claim 1; laser point cloud data at two adjacent moments are obtained by using a laser radar of a differential wheeled robot and the speed and position of the differential wheeled robot are estimated by matching the laser point cloud data at two adjacent moments to obtain a laser odometry B; multiple groups of wheel odometry A and laser odometry B are collected for joint calibration to obtain an optimal rotation matrix and an optimal translation vector, thereby obtaining the required external parameters to complete the joint calibration.

3. A laser radar extrinsic parameter calibration method according to claim 2, characterized in that: The method for matching the laser point cloud data of two adjacent moments is to set corresponding laser point cloud data for the two adjacent frames of laser point cloud data. and ,in Laser point cloud data and The centroid is removed to obtain the corresponding point cloud without the centroid and then perform point cloud registration. The centroid removal method is as follows: the centroids of the two frames of laser point clouds are and , then the corresponding laser point cloud without the centroid is and ; The laser point cloud registration method is: , the matrix Perform SVD decomposition, that is ,in, is a left singular matrix, is a matrix composed of singular values, is a right singular matrix, then the rotation matrix , translation matrix .

4. A laser radar extrinsic parameter calibration method according to claim 3, characterized in that: The joint calibration method is to assume that the pose transformation from the differential wheeled robot base coordinate system to the laser radar coordinate system is the external parameter to be determined. ,exist At time , the wheel odometer reading of the differential wheeled robot is , the laser odometry reading is exist At this moment, the robot's wheel odometer reading is , the laser odometry reading is , the wheel odometer and laser odometer at two moments have the following conversion relationship, Where, A 1,2 express P 1 arrive P 2 The posture change of the wheel odometer, B 1,2 express P 1 arrive P 2. The position change of the laser odometry, combining the above two equations, can be simplified to obtain , solve it using a two-step method: ,in R X To seek external reference The rotation matrix of and are the rotations of transformation matrix A and transformation matrix B respectively, To seek external reference The translation vector of , ,in 、 They are 、 The corresponding Lie algebra can be transformed into: , collect multiple sets of wheel odometer A and laser odometer data B and solve the optimal rotation matrix through least square fitting, which is recorded as R X * , ,Depend on It can be seen that , multiply both sides by ,in I As the unit matrix, rearrange it to solve the optimal translation vector, which is recorded as * ， 。

Images