Proposal of a Pedestrian Recognition Method for Mobile Robots Using Drones Flying Overhead and a Pedestrian Avoidance Navigation Method Based on It

Le résumé fourni par la source

In crowded urban environments, service robots need to guide pedestrians safely while avoiding them. Traditional approaches such as artificial potential field methods enable real-time path planning at low computational cost, but they assume full visibility of pedestrians using on-board sensors such as LiDAR and cameras. In practice, these sensors suffer from occlusion and limited field of view, reducing safety and reliability. To overcome this limitation, we propose a novel pedestrian-aware navigation system that combines a drone-based bird's-eye view with YOLOv8 object detection and artificial potential method. The drone captures overhead images of the environment, from which YOLOv8 detects the positions of pedestrians and mobile robots in real time. The geometric relationship between the drone and the ground is modeled using a pinhole camera model to accurately calculate the coordinates of the pedestrian. These coordinates are transmitted to the robot via ROS, enabling global situational awareness and safe path planning even when on-board sensors fail due to occlusion. Experimental validation, demonstrates that the proposed system accurately detects all pedestrians in the vicinity and enables early and effective collision avoidance. The robot successfully navigates complex scenarios in which pedestrians suddenly appear or do not have direct line-of-sight contact, validating the effectiveness of the proposed drone-assisted navigation system.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.