DO-SLAM: research and application of semantic SLAM system towards dynamic environments based on object detection

Simultaneous Localization and Mapping (SLAM) is one of the research hotspots in the field of robotics, and it is also a prerequisite for autonomous robot navigation. The localization accuracy and stability of traditional SLAM based on static scene assumption declines due to the interference of dynamic objects. To solve the problem, this paper proposes a semantic SLAM system for dynamic environments based on object detection named DO-SLAM. Firstly, DO-SLAM uses YOLOv5 to identify objects in dynamic environments and obtains the semantic information; Then, combined with the outlier detection mechanism proposed in this paper, the dynamic objects are effectively determined and the feature points on the dynamic objects are eliminated; The combination method can reduce the interference of dynamic objects to SLAM, and improve the stability and localization accuracy of the system. At the same time, a static dense point cloud map is constructed for high-level tasks. Finally, the effectiveness of DO-SLAM is verified on the TUM RGB-D dataset. The results show that the Absolute Trajectory Error (ATE) and Relative Pose Error (RPE) are reduced, indicating that DO-SLAM can reduce the interference of dynamic objects.