Research On Multi-sensor Calibration Method In Vehicle-mounted System

Sensor calibration is the process of determining the sensor’s own parameters and pose,including intrinsic calibration and extrinsic calibration.Today’s automated driving technology is increasingly mature,mainly due to the joint use of cameras,inertial measurement elements,multi-beam lidar and other sensing sensors in the vehicle-mounted system,so accurately determine the parameters for the sensor itself as well as the transformation of different sensor coordinate system is a prerequisite to ensure the effective integration of data in unmanned vehicle-mounted systems.At present,in an unmanned vehicle-mounted system containing multiple sensors,the coordinate transformation between the sensors is affected by the calibration results of the intrinsic parameters of the respective sensors,and the sensor parameters provided by the manufacturer are often not very accurate,in addition,when the sensors are used in combination,it is necessary to accurately obtain the corresponding spatial positions between the sensors.In response to these problems,this article first improves the existing intrinsic calibration methods of the camera,inertial measurement unit,and multibeam lidar used in the unmanned vehicle-mounted system to obtain more accurate self-data,and then reasonable combination between the three components and obtain the corresponding accurate sensor extrinsic parameters.Finally,consider the time offset that affects the spatial calibration of the extrinsic parameters to ensure the time synchronization between the sensors.The main work of this subject are as follows:(1)A method for calibrating the internal parameters of the intrinsic measurement unit under the motion model is proposed.The method is to move the inertial measurement unit to different states and stand still to obtain the state information of the accelerometer and gyroscope inside the sensor to obtain the corresponding calibration error.Firstly,the residual error model is constructed according to the working principle of the sensor,and then the Jacobian matrix in the residual error model is solved,and finally the more accurate error parameters are obtained by the iterative optimization method.(2)Introduce the "fragility" concept of lidar point cloud,and improve the calibration method of multi-beam lidar intrinsic parameters.When the lidar extracts the plane information of the calibration target,it adds a distance factor according to the deviation of the laser beam emission,quantify the noise according to the "fragility" of the point cloud near the plane,and fit the point cloud plane to remove the noise on the plane,so as to remove the noise on the plane and optimize the parameters in the Lidar calibration process.(3)A camera-inertial measurement unit external parameter calibration method for moving pixels under B-spline is proposed.This method solves the motion postures of sensors with different sampling frequencies,registers the data collection information of the two sensors at the same time,and adds the concept of "moving pixels" to time synchronization to correct the time offset of the two sensors,ensuring that while the timestamp is synchronized,the corresponding calibration value of the extrinsic space parameter is obtained.(4)Improved and optimized the existing camera-multi-beam lidar extrinsic calibration method containing calibration targets.This method matches the corner points and center points in the point cloud data of the calibration plate collected by the lidar and the image data of the calibration plate collected by the camera,constructs the corresponding objective function,optimizes and solves the extrinsic parameters between the camera and the lidar,and improves the accuracy of the calibration.