A Multi-Object Tracking Method in Moving UAV Based on IoU Matching

Multi-object tracking (MOT) is widely applied in the field of computer vision. However, MOT from a drone's perspective poses several challenging issues, such as small object size, large displacements of targets, and irregular motion of the platform itself. In this paper, we propose an multi-object tracking method based on IoU matching that combines traditional object detection techniques with IoU matching algorithms to achieve simple and fast tracking of targets from a drone's perspective. Unlike other methods, this approach does not require feature extraction for each target but relies solely on the targets' motion information to track, significantly reducing computation time. Additionally, it uses virtual observations to estimate trajectories during target loss and introduces a motion compensation strategy for the drone to reduce error accumulation in filter parameters. Experiments on the Visdrone and UAVDT datasets demonstrate that the proposed method significantly improves performance compared to state-of-the-art drone video tracking methods.