YDM-SLAM: YOLOv8-Powered Dynamic Mapping of Environment Using ORB-SLAM3

Abstract

Simultaneous Localization and Mapping (SLAM) systems enable intelligent navigation for mobile robots. While numerous SLAM systems have been developed and shown success in static environments, they often struggle to handle dynamic environments effectively. In this research, YDM -SLAM is proposed, specifically designed to handle dynamic environments. Built upon the foundation of ORB-SLAM3, YDM-SLAM inte-grates YOLOv8-Seg models for semantic mask generation and real-time removal of feature keypoints from dynamic objects. This integration enables YDM-SLAM to effectively mitigate the impact of moving objects on localization precision. Experimental results on the high dynamic sequence of the TUM-RGBD dataset demonstrate a significant improvement of 78.80 % in terms of absolute trajectory error, highlighting the effectiveness of YDM-SLAM compared to the original ORB-SLAM3. The experiments were conducted using a standard hardware setup, underscoring the practicality and efficiency of the proposed approach in real-world scenarios.