Adaptive mobile camera tracking using color and topological features

Abstract

Various features may be used for tracking with a mobile camera, such as color, edge, contour, texture, locomotion, key point and so on. It has been proved that combination of multiple features can increase the robustness of a tracking system. By considering the characteristics of these features, we propose the approach of employing the color and topological properties of the target in tracking. First, the target is segmented into fragments. The color of the target is represented by a Gaussian Mixture Model(GMM), with every fragment corresponding to a Gaussian model. The topology of the target is represented by a tree-liked structure constructed from the spatial information of the fragments. During the tracking procedure, we extract the foreground by comparing the GMM of the target, and represent it using a GMM as well. After that, color and topological properties of the target are integrated in order to determine the association of the target's fragments with the foreground's fragments. The tracking result is the fragments of the foreground, which are associated with the target's fragments. At last, the color and topological properties of the target are updated by the tracking result. Experimental results demonstrate the effectiveness of this algorithm, even though when the locomotion of the target is unpredictable.