A Global-Local Tracking Framework Driven by Both Motion and Appearance for Infrared Anti-UAV
Yifan Li, Dian Yuan, Meng Sun, Hongyu Wang, Xiaotao Liu, Jing Liu · 2023
Unmanned aerial vehicles (UAVs) have been widely used in various application domains, but unauthorized UAVs may pose a threat to public safety due to violation of aviation regulations. Therefore, how to design an effective UAV tracking method for anti-UAV is a crucial part of the UAV-defense system. In this paper, we propose a Global-Local Tracking Framework driven by both Motion and Appearance (GLTF-MA) including four modules to deal with the practical difficulties in infrared anti-UAV. Firstly, a Periodic Global Detection (PGD) module is periodically performed to re-locate UAVs in the whole image to account for frequent appearance/disappearance and unstable flight paths of UAVs. Meanwhile, a Multi-stage Local Tracking (MLT) module containing a priori stage switching mechanism, motion-appearance matching mechanism, and a motion estimation punisher is routinely implemented to deal with the tiny size of UAVs and background interference. Next, a Target Disappearance Judgement (TDJ) module is performed to give a robust target disappearance flag, followed by a Bounding Box Refinement (BBR) module to refine the target box when the TDJ module thinks the target exists. Extensive experiments demonstrate the superiority of GLTF-MA over other competing counterparts, especially when the UAV is low resolution and moves quickly.