Adaptive edge-based mean shift for drastic change gray target tracking

Numerous tracking approaches have been proposed recently, however, existing approaches can hardly be applied to drastic change gray target tracking because of their vulnerability to unsmooth motion and appearance discontinuity caused by drastic change gray target. We aim to address this problem by proposing an adaptive edge-based mean shift algorithm, where edges are utilized as feature rather than intensity. We propose a method to construct the feature space by employing a combination of edge images to overcome the inadequate representation of single edge feature space. The feature space provides sufficient information for representing targets. In addition, a detection module is proposed to cope with targets whose gray value distributions change drastically. The detection module can detect the target in a specific region which is automatically adjusted according to the target size of former frame; therefore it offers our algorithm the ability to obtain precise target region and target position. By automatically updating the tracker using currently detected target information, our algorithm becomes adaptive to the changes in target's size and shape. Intensive experiments show that our algorithm outperforms state-of-the-art algorithms in tracking drastic change gray targets.