Tracking non-rigid target via dynamic discriminative geodesic active contours

In this paper, we propose a novel dynamic framework for tracking non-rigid target in video sequence. To solve the problems caused by various situation and sophisticated background during tracking of a target, a discriminative appearance model based on dynamic SVMs is first proposed to extract the rough target region, which provides discriminative target region information clues. In addition, we design a new discriminative geodesic active contour model (DGACM) that combines both edge and discriminative region information, thereby enabling the extraction of more precise object boundaries compared to conventional active contour models. By adjusting the proportion between edge and region constraint terms, the proposed model can adapt to appearance variation, sophisticated background and occlusion. Experiments on synthesized and real image sequences demonstrate that the proposed method achieves more robust and effective deformable target tracking in challenging situations compared to other competitive contour tracking methods.