Robust tracking and lost target re-acquisition in video sequences using a combined color-gradient orientations based particle filter and covariance matching

Making visual features/trackers cooperate together allows to benefit from the complementary of different features and explores the merits of each individual tracker, which increases the robustness of video tracking. In this paper, two trackers are made in cooperation. The main tracker is based on particle filter and uses a visual model combining color and gradient orientations as the target representation, along with a conditional updating strategy. By using a probabilistic search, this tracker has the advantage of reducing the search space by only generating target hypothesis around the last estimated location. However, it may drift due to unexpected motion, occlusions and out of view problems. For this, covariance matching is used as an auxiliary tracker to assist the former search mechanism. By adopting an exhaustive search, covariance matching takes the control when the main tracker fails in locating the target. Once the target is re-detected, the tracking control is returned back to the probabilistic search based tracker. Many experiments show that the proposed tracker exhibits competitive tracking results when compared with other approaches. In addition, it allows tracking through occlusions and has the capability of re-detecting a target after disappearing and appearing again in a video sequence.