AD-YOLO: A unified method for traffic dense and small object detection in UAV images

The densely distributed, scale-varying objects in unmanned aerial vehicle (UAV) images, together with dynamic, diverse, and unconstrained backgrounds, make conventional detection methods prone to missed detections, false alarms, and localization biases. To empower UAV vision tasks, we propose AD-YOLO, a unified framework tailored for traffic dense and small objects in drone imagery. First, a module, combining adaptive rotation convolution unit and grouped directional attention with mixed-kernel, is introduced to enhance the orientation invariance and multi-scale discrimination. Then, a dual-path collaborative feature pyramid network is proposed to jointly refine semantic and spatial details via multi-directional context aggregation and hierarchical spatial preservation flows. Last, a hierarchically dense reparameterized large-kernel module is designed to achieve broader receptive fields with reduced computational complexity. Extensive experiments on the VisDrone2019 and UAVDT datasets demonstrate that AD-YOLO outperforms state-of-the-art methods in detection accuracy while maintaining favorable computational efficiency, highlighting strong potential of real-life traffic monitoring applications.