PLRUT: Pseudo Label and Re-detection Boosted Unsupervised Tracking of Unmanned Aerial Vehicle Objects

Jun Wang, Huadong Dai, Bo Zhang, Shan Qin, Jian Zhao, Zhihao Zhang

Abstract: In recent years, visual object tracking technology has seen significant advancements. Nevertheless, the tracking challenge for mobile platforms, such as unmanned aerial vehicles (UAVs), remains a relatively unexplored domain. Hence, this paper introduces a Pseudo Label and Re-detection boosted Unsupervised Tracking of Unmanned Aerial Vehicle Objects (PLRUT) approach, aiming to effectively leverage unlabeled data for exploration. We introduce an innovative strategy to generate pseudo labels for unsupervised training, addressing the challenge of insufficient pseudo label generation in previous methods when operating in mobile environments. Additionally, we introduce a global re-detection mechanism to mitigate target disappearance during motion. Experimental results demonstrate that our proposed tracking and detection approach can effectively achieve UAV tracking, outperforming previous unsupervised methods.