Visual SLAM Algorithm Based on YOLOv5 in Dynamic Scenario

Aiming at the problem of inaccurate pose estimation caused by dynamic feature points produced by dynamic objects during the operation of traditional SLAM system, a visual SLAM algorithm based on YOLOv5 is proposed to identify environmental images, distinguish dynamic objects in the scene and eliminate their dynamic feature points, and then use the remaining static feature points to perform oppositional geometric constraints. Finally determine the pose transformation matrix to improve the accuracy of the system. In this paper, the published TUM dataset is used to carry out experiments, and the results show that the algorithm adopted in this paper improves the relative positioning results of SLAM system in common dynamic situations, and reduces the mean of the median range of relative position error (RPE) in the monocular mode of the original ORB-SLAM3 by 91.53%. Effectively improve the positioning accuracy of the system.