Drone-Based Environment Perception for Manually Operated Industrial Trucks

Abstract

Manually operated industrial trucks remain indispensable in intralogistics but are prone to accidents caused by limited operator visibility, particularly at intersections and when handling large loads. Existing operator assistance systems are vehicle-mounted or otherwise limited to a direct line of sight, and none provide an elevated, forward-looking monitoring perspective. This paper presents the concept and initial results of a drone-based operator assistance system for the anticipatory environmental monitoring of manually operated forklifts, in which a drone equipped with a fisheye camera flies ahead of the forklift to detect lanes, estimate its pose, and identify hazards. The concept is implemented on an experimental setup using a stationary sensor package in place of the drone. The results demonstrate promising wireless network performance with regard to latency requirements, as well as promising lane detection with an mAP@50 value of 0.986 and person detection with a precision value of 95.2% using YOLO-based networks. The latter also reveals challenges in more crowded scenes. Marker-based pose estimation further enables a correctly oriented representation of the surroundings on a schematic bird’s-eye-view map, achieving a maximum positional deviation of 5.0 cm and a rotational deviation of 4.3°. Validation with an actual, moving drone as well as the acquisition of larger, more diverse datasets remain subject to future work.