Lidar Odometry Key Frame Selection Based on Displacement Vector Similarity
作者单位：School of Mechatronic Engineering and Automation Shanghai University
会议名称：《第40届中国控制会议》
会议日期：2021年
学科分类：080904[工学-电磁场与微波技术] 0810[工学-信息与通信工程] 0809[工学-电子科学与技术（可授工学、理学学位）] 08[工学] 081105[工学-导航、制导与控制] 081001[工学-通信与信息系统] 081002[工学-信号与信息处理] 0825[工学-航空宇航科学与技术] 0811[工学-控制科学与工程]
关 键 词：Lidar Odometry Cosine Similarity Normal Distributions Transform(NDT)
摘 要：In order to improve the efficiency and reduce the cumulative error of lidar odometry caused by key frame update. This paper proposes a key frame selection method based on displacement vector similarity. First, calculate the pose between the current frame and the reference key frame according to the Normal Distribution Transform（NDT） algorithm;then, set the Manhattan distance between the two frames above, and calculate the cosine similarity of the displacement vector;finally, update the key frame according to the similarity threshold. The experimental results show that the key frame selection method proposed in this paper can effectively improve the key frame update efficiency and reduce the cumulative error of the odometry: compared with the NDT method updating key frame based on the distance, the maximum absolute error is reduced by 11.88 meters, and the relative error is reduced by 8%,time consumption has been reduced by 55 seconds.