Research on SLAM Dynamic Target Removal Method Based on Convolutional Neural Network

Integrating object detection into the ORB-SLAM2 system, using object detection algorithms to annotate objects with motion attributes in the environment, and then relying solely on the keyframes of static objects for map construction, can effectively remove the impact of dynamic objects on semantic map construction. This article uses YOLO4 for dynamic object detection in SLAM system, and uses this object detection algorithm to remove dynamic objects from environment. Simulation experiment shows that 3D semantic map effect of dynamic object removal is good. It is of great significance to increase perception ability of mobile robots to environment.