Research and Application Analysis of Visual SLAM Algorithm Integrated with YOLOv5 in Dynamic Environments

The paper presents a visual simultaneous localization and mapping (SLAM) system enhanced by integrating YOLOv5 for dynamic environments. Traditional visual SLAM systems often struggle with positioning accuracy and robustness when dynamic objects are present, as these systems typically rely on static feature points. The proposed method leverages YOLOv5 for object detection to identify and classify dynamic elements, such as people and vehicles, and eliminates their feature points, thereby reducing the influence of dynamic objects on the SLAM system. The integration is achieved by modifying the ORB-SLAM3 framework, specifically enhancing its tracking thread with YOLOv5's dynamic feature point elimination. Experimental validation on the TUM dataset demonstrates a significant improvement in attitude estimation accuracy, particularly in highly dynamic scenarios, with performance enhancements exceeding 90% compared to traditional methods. The results highlight the enhanced adaptability and robustness of the improved SLAM system in complex environments, addressing the challenges posed by dynamic objects and optimizing its utility for applications in indoor positioning, autonomous driving, and mobile robotics.