U-Tracker: efficient multiobject tracking for drone vision

Multi-object tracking (MOT) plays a crucial role in many fields, with its application in the drone domain being particularly significant. With efficient MOT methods, drones can accurately identify and track multiple targets in complex aerial environments, greatly enhancing their performance in military reconnaissance, border patrol, disaster relief, and other tasks. Existing drone-based MOT methods often fail to effectively utilize temporal information during detection, which impacts tracking accuracy. To address this issue, this paper proposes an efficient decoder for utilizing temporal information in images: U-Tracker. We validate the effectiveness of our method on several well-known datasets, such as VisDrone2019 and UAVDT.