Collaborative Low Frame Rate UAV Tracking by Proposals

Yong Wang, Jiaqi Zhou, Juntao Liang, Xiangyu Zhu, Zhoujingzi Qiu

Abstract: In this letter, we present a framework for low frame rate tracking from unmanned aerial vehicles (UAV). This is a challenging task in which the target has to be localized in a large search space. This problem is amplified by the fact that the tracked object is most often composed of a limited number of pixels which makes it difficult to distinguish from the background. Our method consists of three components: a tracker, a proposal provider, and a coordinator. The tracker aims at providing effective tracking inference. The proposal provider checks the tracking results and provides candidates when needed. The coordinator module adjusts the tracking according to the target response. With such collaboration, our method provides both tracking efficiency and reliable target localization. Two UAV datasets are down sampled to 3 frame per second (fps) for validation. We demonstrate with extensive experimental results that this framework can significantly enhance the tracking performance. Moreover, as a general framework, our approach is very flexible and can thus be extended to other applications.