Multiscale Space Adaptation-Based Dynamic Object Removal for LiDAR Mapping

Simultaneous Localization And Mapping (SLAM) is a crucial technology for mobile robots to locate themselves when working in a location environment. However, when a large number of dynamic objects appear in the environment, the positioning and mapping results of the robot will be affected. At present, online real-time removal algorithms can effectively solve such problems. Therefore, this paper proposes a dynamic object removal algorithm based on multi-scale spatial information, which identifies and eliminates moving objects in the environment during self-localization and environmental modeling. In this paper, the Inertial Measurement Unit (IMU) data is used to pre-liminarily judge the robot's position and attitude transformation, and the dynamic target distance is combined to adjust the target feature extraction scale, and modify the dynamic object detection conditions to screen objects with high confidence. To remove misjudgment information caused by irregular objects in the environment, road information is used to remove misjudgment information and obtain dynamic object coordinate information in the environment. Through quantitative and qualitative experiments, it has been proven that this algorithm can improve the positioning accuracy of robots in urban road scenes and optimize environmental models. This algorithm reduces Absolute Pose Error by 0.11032m.