Image-based Pedestrian Detection and LiDAR Data Integrated Path Planning for Mobile Robots
Sheng-Ru Chen, Sung-Hua Chen · 2025
This paper proposes a sensor fusion method based on Dist-YOLO and LiDAR to enhance the performance of the Dynamic Window Approach (DWA) in dynamic obstacle avoidance, referred to as YOLO-Dynamic Window Approach (YDWA). The proposed method employs the Dist-YOLO model to detect pedestrians and obtain their distance and position information, subsequently integrated with LiDAR data to compensate for LiDAR’s limitations in detecting distant or occluded objects. Specifically, for obstacles detected by LiDAR, a Kalman filter is employed to predict the position and velocity of dynamic obstacles, further enhancing the predictive capability in close-range environments. The DWA generates and selects the optimal path by evaluating the cost of various velocity and steering candidates. Based on this principle, this paper improves the cost function of conventional DWA by incorporating pedestrian detection and dynamic obstacle information. This enhancement enables the path planning algorithm to predict and avoid pedestrians or dynamic obstacles in advance, effectively improving both safety and stability in robot navigation.