Self-Adaptive Visual Tracker Based on Background Information

Occlusion is a thorny issue in visual tracking, which may lead to serious drift to the tracking result. In this paper, a new tracker is proposed to deal with occlusion in tracking. Background surroundings to the object are divided into patches as supplementary information for occlusion detection. When the object is partially occluded, compensation will be made to the estimated position thus ensuring a better tracking result. The detector will be activated to search the whole frame for the object when object is missing. Correlation filter is applied to build the classifier for a higher speed and random ferns are used in the detector. Experiments are carried out on OTB benchmark videos and the result indicates the proposed tracker is preferable comparing to state-of-art trackers in handling occlusion.