基于LM算法的运动相机与激光雷达联合标定方法  

Joint calibration of sports camera and lidar based on LM algorithm

作　　者：李红帅 罗笑南 邓春贵 钟艳如[2] LI Hongshuai;LUO Xiaonan;DENG Chungui;ZHONG Yanru(School of Computer and Information Security,Guilin University of Electronic Technology,Guilin 541004,China;Guangxi Key Laboratory of Image Graphics and Intelligent Processing,Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学计算机与信息安全学院,广西桂林541004 [2]桂林电子科技大学广西图像图形与智能处理重点实验室,广西桂林541004

出　　处：《桂林电子科技大学学报》2022年第5期345-353,共9页Journal of Guilin University of Electronic Technology

基　　金：国家自然科学基金(62033001);广西科技创新重点项目(AA18118039-4);广西图像图形与智能处理重点实验室基金(GIIP201703)。

摘　　要：为解决运动相机与激光雷达之间的数据匹配问题,设计了一种基于Levenberg-Marquard(LM)算法的运动相机与激光雷达联合标定优化方法。首先将标定板放置在激光雷达与运动相机公共视野范围内,通过改变标定板位置,采集不同位置下标定对象的激光点云和图像数据。然后通过OpenCV调用鱼眼畸变校正函数对图像畸变校正,并获取多组标定板图像角点像素坐标。与此同时对激光点云进行点云滤波和点云配准操作,并采用手动与自动相结合的方法对激光点云进行分割,进而通过点云中心迭代算法求解出标定板点云中心坐标和各个角点的点云坐标。最后通过多组表示标定板角点的点云坐标和相对应的图像像素坐标利用直接线性变换法(DLT)计算两传感器间联合标定初值,并构造点云重投影坐标与图像像素坐标差值的最小二乘函数,通过引入阻尼因子的LM算法对该函数进行优化,并求解出优化后的联合标定结果。实验结果表明,联合标定结果与初值相比,重投影误差降低了35%,利用联合标定结果基于共线方程原理实现激光点云与图像融合,验证了该方法的准确性和有效性。In order to solve the data matching problem between sports camera and lidar,a joint calibration optimization method of sports camera and lidar based on Levenberg-Marquard(LM)algorithm is designed.First,the calibration board is placed in the common field of view of the lidar and the sports camera,and the laser point cloud and image data of the calibration object at different positions are collected by changing the position of the calibration board.Then the fisheye distortion correction function is called through OpenCV to correct the image distortion,and obtain multiple sets of pixel coordinates of the corner points of the calibration plate image.At the same time,point cloud filtering and point cloud registration are performed on the laser point cloud,and the laser point cloud is segmented by a combination of manual and automatic methods,and then the point cloud center iterative algorithm is used to solve the calibration board point cloud center coordinates and The point cloud coordinates of each corner point.Finally,through multiple sets of point cloud coordinates representing the corner points of the calibration board and the corresponding image pixel coordinates,the direct linear transformation method(DLT)is used to calculate the initial value of the joint calibration between the two sensors,and the difference between the point cloud reprojection coordinates and the image pixel coordinates is constructed.The least squares function of,the function is optimized by the LM algorithm that introduces the damping factor,and the optimized joint calibration result is solved.Experiments show that the joint calibration result reduces the reprojection error by 35%compared with the initial value.The joint calibration result is used to achieve laser point cloud and image fusion based on the principle of collinear equations,which verifies the accuracy and effectiveness of the method.

关 键 词：激光雷达 运动相机 联合标定 图像畸变校正 直接线性变换法 最小二乘法 LM算法 OPENCV 

分 类 号：TN958.98[电子电信—信号与信息处理] TP391.4[电子电信—信息与通信工程]