Decoupling Ego-Motion from Target Dynamics via Dual-Interval Motion Cues for UAV Detection

This research addresses challenges in object detection from Unmanned Aerial Vehicles (UAVs) caused by severe ego-motion, camera jitter, and scale variations. The authors propose a vision-only motion-guided detection framework that decouples target motion from camera-induced disturbances. Key components include Homography-based Global Motion Compensation (GMC) to align adjacent frames and a Dual-Interval Motion Extraction strategy capturing short-term and long-term motion cues. A lightweight Dual-Interval Motion Injection (DIMI) module integrates these cues within a Feature Pyramid Network. Experiments on the VisDrone-VID dataset show consistent improvements over the YOLOv8n baseline under severe ego-motion, with ablation studies confirming the effectiveness of the proposed design.