Most existing simultaneous localization and mapping(SLAM)methods perform stably under ideal conditions,but the localization accuracy decreases in dynamic environment due to mismatches of feature point clouds caused by moving objects.To address this issue,a dynamic point cloud detection algorithm is proposed.Firstly,the inertial measurement unit information is used to preprocess the point cloud data,including operations such as distortion removal.Secondly,the ground point clouds are eliminated,and the non-ground point clouds are clustered by using the curvature voxel structure.Thirdly,the Hungarian algorithm is used to associate and match the clusters between the two frames,while unifying the coordinate system with inertial information.Finally,a screening strategy is designed:coarse screening of dynamic clusters using the intersection-over-union of bounding boxes and centroid velocity,followed by fine screening based on the similarity of point cloud distribution along the z-axis(height).Experimental results show that the proposed algorithm can effectively identify and filter out most dynamic point cloud clusters in the environment.Compared to the LIO-SAM algorithm,the root mean square error of localization in the four scenarios is reduced by 17.75%on average.The average accuracy and recall rate are 14.81%and 5.90%higher than Removert,respectively.
关键词
激光同步定位与建图/动态环境/动态点云检测/筛选策略
Key words
LiDAR simultaneous localization and mapping/dynamic environment/dynamic point cloud detection/filtering strategy
维普
万方数据