Transformer Enhancer for UAV Tracking

Most previous progress in object tracking is realized in daytime scenes with favorable illumination. State-of- the-arts can hardly carry on their superiority at night so far, thereby considerably blocking the broadening of visual tracking-related unmanned aerial vehicle (UAV) applications. To realize reliable UAV tracking at night, this work introduces the Transformer to the low-light enhancement and develops a brand new spatial-channel Transformer-based enhancer, namely SCT, which is trained in a novel tracking-specific manner and plugged prior to tracking approaches. To exploit global information that is crucial in low-light enhancement, the novel spatial-channel attention module is proposed to model long-range inter-independencies while preserving local context. In the enhancement process, SCT relights low-illumination images and alleviate noise through a robust non-linear curve projection. Moreover, to provide a comprehensive evaluation, we construct a large scale night tracking benchmark, namely DarkTrack2021, which contains 110 challenging sequences with 100K frames in total. Evaluations on both the public UAVDark135 benchmark and the newly constructed Dark- Track2021 benchmark show that the tracking-specific design enables SCT with significant performance gains for nighttime UAV tracking compared to other top-ranked low-light enhancers. Real-world tests on a typical UAV platform further verify the practicability of the proposed approach.

Guangze Zheng
Guangze Zheng
Research assistant in Vision4Robotics

My research interests include deep learning, visual object tracking, and robotics.