TAE: Target-aware enhancer for nighttime UAV tracking

A new framework called TAE has been proposed to enhance low-light image quality for nighttime UAV tracking, addressing the challenges of severe image degradation that hinder effective object tracking in dark conditions. TAE utilizes weak supervisory signals from tracking bounding boxes to focus enhancement efforts on target areas, while employing an adaptive RGB multi-curve fusion mechanism for improved modeling across different regions.

This development is significant as it not only improves the accuracy of UAV-based tracking systems in low-light environments but also contributes to the establishment of DarkSOT, a benchmark dataset that includes 268 sequences for further research in nighttime UAV tracking, potentially advancing applications in various fields such as surveillance and search and rescue operations.