Approach to 3D SLAM for mobile robots based on point-line features and superpixel segmentation in dynamic environments

This paper proposes a point-line features based 3D SLAM algorithm tailored for mobile robots operating in dynamic scenes. First, the line features are optimized by a merging strategy that combines the scale hierarchy, angle, midpoint spacing, and endpoint spacing of the line segments. Subsequently, a dynamic object removal approach coupling Delaunay triangulation and superpixel segmentation is presented. Also, a high-quality keyframe selection strategy is constructed by evaluating the robot's motion magnitude, the proportion of dynamic pixel points, and the success rate of keyframe tracking. On this basis, a robust loop closure detection method is developed by incorporating point-line features. Experimental results show that the proposed algorithm significantly improves localization accuracy and map consistency in dynamic scenes. Compared with ORB-SLAM2 and DS-SLAM, the absolute trajectory accuracy in high-dynamic scenes have been improved by an average of 96.03% and 26%, respectively, and its robustness is superior to existing state-of-the-art methods.