Object Detection in UAV Images Based on RT-DETR with CG Downsampling and CCFMP
Chushi Yu, Yoan Shin · 2024
In recent years, object detection using unmanned aerial vehicle (UAV) images has emerged as a crucial task due to the advancement in drone technology. However, detecting targets in UAV images pose challenges such as difference size, changing shapes, occlusions, and lighting condition. Despite the commendable results achieved by object detection algorithms based on deep learning neural networks, they still encounter numerous missed detections and false alarms. In order to solve the problems of complex scenes, diverse sizes, dense small targets and severe occlusion, we proposed an enhanced object detection method based on real-time detection transformer (RT-DETR) with context guided downsampling in UAV images. Numerical experiments conducted on the VisDrone dataset demonstrate the effectiveness of the proposed method, improving detection accuracy and model's robustness and capability in complex environments.