Visual SLAM in Dynamic Environment Based on ORB-SLAM3

At present, the mainstream SLAM system mainly works in the static environment, and there will be a large error in the dynamic environment. This paper presents a VSLAM method for dynamic feature point detection based on target detection algorithm and LK optical flow method. The SLAM system is capable of robust, accurate and continuous operation in highly dynamic environments. The method first obtains the potential dynamic frame by passing the image into the target detection network. Then the matching feature points of two adjacent frames are obtained by optical flow method. In order to accurately remove dynamic feature points, the dynamic frame is first identified by the threshold value of the dynamic frame, and then the dynamic feature points in the dynamic frame are identified according to the threshold value of the dynamic feature points. Finally, the dynamic feature points in the dynamic frame are eliminated. Our proposed algorithm has been tested on TUM dataset for many times, and the performance is significantly improved compared to the original ORB-SLAM3. Compared with the current better methods, this method can significantly improve some indexes.