Semantic SLAM based on ORB-SLAM3 combined with YOLOv5 in Dynamic Environment

At present, the technology of intelligent mobile robots is developing rapidly, and Simultaneous Location and Mapping (SLAM) is one of the core technologies for robot intelligence. This technology enables robots to achieve autonomous positioning in unknown environments and simultaneously generate maps of the surrounding environment, greatly expanding the application fields of robots. The traditional visual slam system is susceptible to dynamic interference (such as people and vehicles) in dynamic scenes, which leads to the decline of positioning accuracy, the "ghost" of map building and the lack of system robustness. To solve this problem, this paper proposes an dynamic visual slam system based on semantic recognition(yolo-v5), which improves the system performance through semantic segmentation and point cloud optimization.