UAV Target Tracking in Complex Nighttime Environments

Currently, the target tracking of unmanned aerial vehicles (UAVs) in complex nighttime environments faces numerous challenges. Existing methods have deficiencies in low-light image enhancement and robust tracking. They struggle to make full use of the multi-domain features of images and fail to process different brightness regions in low-light images precisely enough. To address these issues, a joint paradigm of joint spatial and frequency domain feature diffusion learning (JSFDF) is proposed. This paradigm can fully utilize the features in both the spatial domain and the frequency domain, effectively achieving the enhancement of low-light images. At the same time, a prior contrast method based on the dark channel and bright channel is constructed. This method guides the model to focus on the low-brightness pixel parts of low-light images and the regions with strong contrast between the high-brightness pixels of the enhanced images, thereby achieving robust nighttime UAV tracking. The constructed tracker is named PCDB. Through experiments conducted on the NAT2021 and darktrack2021 datasets, the results show that, compared with the existing nighttime UAV target tracking methods, the method proposed in this paper demonstrates excellent performance.