Vision-Based Tracking and Following of a Moving Target Using an Unmanned Aerial Vehicle

Abstract

This work introduces an autonomous system for mobile target tracking and follow ing using vision-based (RGB) uni-modal data, specifically designed for unmanned aerial vehicles (UAVs) and enhanced by multi-target information. It addresses the gap in current research by applying state-of-the-art multi-object tracking (MOT) techniques to target following scenarios, moving beyond traditional single-object tracking (SOT) methods. The system combines the real-time object detector YOLOv8 with MOT algorithms BoT-SORT and ByteTrack to extract and uti lize multi-target data, improving re-identification performance and reducing ID switches, especially under partial or full occlusions in dynamic environments. A 3D flight control mechanism is implemented to enable responsive target following, maintaining line-of-sight despite changes in target speed or direction. The system is validated through simulation testing, demonstrating accurate and robust track ing that effectively differentiates the intended target from surrounding bystanders. By tackling key challenges, this work paves the way for practical UAV applications in vision-based target following using multi-target information.

9