Dynamic Scene SLAM Algorithm Based on Target Detection and K-Means Clustering

In order to solve the problem that the camera cannot be accurately localised due to the interference of dynamic objects in the environment, an RGBD SLAM algorithm for dynamic scenes is proposed, which combines the target detection algorithm with the K-means clustering algorithm to separate the dynamic and static feature points in the environment, and is able to remove the dynamic feature points while retaining the static background points of better quality. The algorithm is simulated in the TUM dynamic Simulation experiments were carried out in the TUM dynamic sequence dataset, and the experiments show that the algorithm has good positioning and tracking effect in both high and low dynamic scenes, especially in the high dynamic scene, compared with the traditional ORBSLAM2 algorithm, the camera's absolute trajectory rms error decreases from 94.2% to 95.96%, which substantially improves the positioning accuracy and robustness of the camera's position estimation in the high dynamic scene.