Robust Vision SLAM Based on YOLOX for Dynamic Environments

Simultaneous localization and mapping (SLAM) is used in almost all intelligent robots. Recently, many excellent SLAM systems have been developed with good performance and impressive ideas. However, there are many problems that do not have a good solution. For example, when dynamic objects in the system are so perturbed that the system is unstable. This paper proposes a visual SLAM in a dynamic environment, called YOLOX-SLAM, which has good robustness. In YOLOX-SLAM, four threads run in parallel: tracking, object detection, local mapping, and loop closure. YOLOX-SLAM uses YOLO object detection to detect dynamic objects and eliminate dynamic feature points. Good accuracy is achieved in a dynamic environment, reducing the influence of dynamic objects. Our experiments were validated on the TUM RGB-D dataset. The results show that YOLOX-SLAM has better accuracy than ORB-SLAM2 in highly dynamic scenarios.