Visual Object Tracking in The Dark Videos with Siamese Networks

In the past few years, visual object tracking has achieved significant results. Despite having a great performance, there are still so many challenges in this field. Low-light videos are one of those challenges. Most visual object trackers are trained in a way to have their best performances in normal light conditions. Also, the majority of applications that are using visual object tracking, require it to be able to perform in real-time. In this research, we use state-of-art and fast enhancers to make the illumination of low-light videos, closer to normal conditions. Then, we check their effect on a visual object tracker called, SiamFC++. After that, to increase the speed of enhancers, we propose the method of rescaling. Then, we check the effect of rescaling on the proposed approaches. To evaluate the proposed approaches, we use the videos of a public dataset called VAVDark135 which is generated to evaluate visual object trackers in low-light conditions. The results of this research are so promising.