Nighttime Aerial Tracking

Overview

Exploit low-light enhancement, denoising, domain adaption technologies to adapt SOTA tracking approaches to low-light conditions. With slight computational consumption, enable UAV tracking at night.

Papers with code

Related works are presented as follows:

• Designed a Retinex-inspired plug-and-play deep low-light enhancer, dubbed DarkLighter, to light up the darkness for UAV tracking. Experiments on a dark tracking benchmark verify its effectiveness in several trackers and superiority against other SOTA general low-light enhancement algorithms, with sufficient real-time speed on an embedded system.

  DarkLighter: Light up the Darkness for UAV Tracking in IROS 2021

• Constructed a spatial-channel Transformer (SCT) enhancer to facilitate nighttime UAV tracking in a task-inspired manner. Evaluations on the public UAVDark135 and the newly constructed DarkTrack2021 benchmarks demonstrate that the performance gains of SCT brought to nighttime UAV tracking surpass general low-light enhancers.

  Tracker Meets Night: A Transformer Enhancer for UAV Tracking in RA-L with ICRA2022 presentation

• Proposed an unsupervised domain adaptation framework to adapt object tracking from daytime to nighttime, along with a nighttime tracking benchmark.

  Unsupervised Domain Adaptation for Nighttime Aerial Tracking in CVPR2022

Benchmarks

We construct some pioneer benchmarks to serve for the development of nighttime object tracking: