Research on Visual SLAM Localization of Mobile Robots in Dynamic Scenes Based on YOLOv8m-seg

The rapid advancement of intelligent mobile robots has led to significant attention on SLAM technology in various domains. However, traditional SLAM systems are typically designed with the assumption of a static environment and often struggle with accuracy and robustness when applied to dynamic settings. To overcome this challenge, this paper introduces a novel visual SLAM algorithm, named PL YL-SLAM, based on YOLOv8m-seg, which aims to enhance localization accuracy and robustness in dynamic environments. The algorithm is built on the ORB-SLAM3 framework, incorporating the LSD line segment detection method into the tracking module. Furthermore, a dynamic object detection thread is introduced to handle moving regions during camera motion. This thread uses the YOLOv8m-seg instance segmentation network to detect dynamic regions, generate dynamic masks with semantic labels, and eliminate point and line features related to moving objects. Experiments using the publicly available TUM dataset confirm that the proposed framework significantly enhances the localization accuracy and robustness of visual SLAM in dynamic environments.