A novel method of object tracking with an adaptive template

Abstract

This paper proposes a novel object tracking approach in terms of adaptive template matching. In an image sequence, image is smoothed and segmented, then its edge is extracted. General distance that is between a pixel of edge template and neighbor set of its matching pixel in image is defined as analogically as definition of partial Hausdoff distance. First, coarse matching is used to find n candidate subimages by means of edge template. Secondly, fine matching is used to locate the optimal matching in n candidates by means of gray region correlation. The heuristic rule of revising edge template is proposed in 8-connection system according to MRF model. This strategy prevents template from drifting towards background. This experimental results present that this algorithm can effectively and efficiently track object in an image sequence under complex scene.