Research on RGB-D SLAM Dynamic Environment Algorithm Based on YOLOv8

RGB-D SLAM mainly uses an RGB-D camera as a sensor to extract feature points and construct maps through RGB images and depth images. However, when faced with dynamic environment, due to the static assumption of SLAM system, the accuracy of the camera Pose estimation will be biased to some extent. Inspired by YOLO’s performance in dynamic object detection, this paper proposes an algorithm based on the combination of YOLOv8 and ORBSLAM2, which uses the location information of dynamic objects detected by YOLOv8 to eliminate feature point within the corresponding areas, aiming at improving the accuracy and robustness of SLAM system positioning in dynamic environment. At the same time keep its own characteristics of real time. The test on TUM open data set shows that compared with the original ORBSLAM, the algorithm improves the positioning accuracy camera pose and has been improved in three directions of xyz, which verifies the feasibility and effectiveness of the algorithm.