Abstract: Visual object tracking is an important branch in computer visions. In recent years, with the remarkable success of deep learning techniques, a series of deep tracking algorithms have emerged with impressive performances. In this paper, we review the recent development of deep learning based trackers. First, we revisit the development of tracking benchmarks in the last decade. These tracking datasets not only comprehensively help evaluate the tracking algorithms but also largely support the model training of deep trackers. Next, we discuss several representative tracking frameworks including deep correlation filter tracking, classification-based tracking networks, Siamese tracking networks, gradient-based tracking networks and Transformer based deep trackers. Finally, we conclude the paper and discuss the potential future research directions of the visual tracking.

Key words: deep visual tracking, benchmark datasets, correlation filter, classification-based tracking networks, Siamese tracking networks, gradient-based tracking networks