Robust SLAM based on Segmentation of Dynamic Object’s Point Cloud

抄録

At present, the mainstream Simultaneous Localization and Mapping(SLAM) system is mainly suitable for static scenes. However, in real life, dynamic objects in an unknown environment limit the application of the SLAM system. Using the traditional SLAM scheme, if the moving object exists in the sensor field of view, the constructed 3D point cloud map leaves the point cloud of the moving target multiple times, and affects the accuracy of the visual odometry(VO) and closure loop detection. This paper proposes a scheme for SLAM in dynamic environment that can be used to identify dynamic objects and remove its point clouds. The robustness of the system in dynamic environment is verified by the outliers of feature points matching and the point cloud map.