Fig. 2

Comparison of the proposed approach against state-of-the-art tracking methods ADNet on three example sequences challenged by large occlusion (top row), illumination variation (middle row), and fast motion (bottom row). Our approach performs more robustly than the ADNet tracking method