A Robust and Fast Object Tracking Method Using a Dynamic Mask and an Adaptive Search

Abstract

This paper proposes a robust and fast object tracking method using a dynamic mask and an adaptive search for monitoring suspicious persons and objects. The dynamic mask removes appearance changes such as deformation and occlusion from a target region for improving robustness of tracking. The adaptive search restricts an area to detect a target using spatial temporal consistency of the target for improving the robustness and processing speed. Two experiments were conducted with a dataset which is open to the public. First experimental results show that our method improves the robustness to the appearance changes by 1.2-2.7% compared with a conventional method. Second experimental results show that our method achieves the processing speed of 83.3 fps, which is approximately 5.9-208.3 times faster than that of conventional methods. With the high robustness and the fast processing speed, a real-time surveillance system with pan-tilt-zoom cameras can be developed for monitoring the suspicious persons and the objects.