先验地图/IMU/LiDAR的图优化和ESKF位姿估计方法对比  

Comparison of prior map/IMU/LiDAR map optimization and error state Kalman filter pose estimation methods

作　　者：郭文卓 李林阳[1,2] 程振豪 杨朝永 张乐添 赵冬青 GUO Wenzhuo;LI Linyang;CHENG Zhenhao;YANG Chaoyong;ZHANG Letian;ZHAO Dongqing(Institute of Geospatial Information,Information Engineering University,Zhengzhou 450001,China;School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]信息工程大学地球空间信息学院,郑州450001 [2]武汉大学测绘学院,武汉430079

出　　处：《测绘科学》2023年第4期88-97,共10页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41774037,42104033);博士后科学基金项目(2022M712442)。

摘　　要：针对滤波和优化融合算法在不同场景下定位性能不明确的问题,该文构建了一种融合先验点云地图、激光雷达(LiDAR)、惯性测量单元(IMU)的位姿估计框架。对比分析了基于图优化和误差状态卡尔曼滤波(ESKF)两种算法的位姿估计精度,并采用3组KITTI数据进行实验分析。结果表明:图优化算法的绝对位姿误差的均方根小于ESKF算法,3组数据的精度分别提升了28.9%、12.5%和21%;在复杂场景下,基于图优化算法的性能高于滤波算法;在简单场景下,滤波和图优化算法的精度接近,而滤波算法更加稳定。Aiming at the problem that the positioning performance of filtering and optimization fusion algorithm is not clear in different scenes,a pose estimation framework integrating prior point cloud map,light detection and ranging(LiDAR) and inertial measurement unit(IMU) was constructed in this paper.The pose estimation accuracy based on graph optimization and error state Kalman filter(ESKF) algorithms was compared and analyzed,and three groups of KITTI data were used for experimental analysis.The results showed that the absolute root mean square error of the graph optimization algorithm was less than that of the ESKF algorithm,and the accuracy of the three groups of data was improved by 28.9%,12.5% and 21% respectively;in complex scenes,the performance of graph based optimization algorithm was higher than that of filtering algorithm;in simple scenes,the precision of filtering and algorithm was close and the filtering algorithm was more stable.

关 键 词：先验地图 误差卡尔曼滤波 图优化 位姿估计 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]