Visual object tracking with the adaptive scale based on AGMM point sets matching

A visual object tracking method with the adaptive scale based on AGMM(Asymmetrical Gauss Mixture Models) point sets matching is proposed aimed at adaptively following the object's scale changes, which often cause tracking failure. As the feature point set in the last frame is considered as the GMM centroids and the feature point set in the current frame represents the data respectively, AGMM fuses the feature information and spatial information; by comparing the similarity between data and GMM centroids, we match the point sets between two adjacent frames and obtain the reliable feature points in the current frame; the degree of dispersion between points in the point set accurately reflects the size of the object scale and by using affine transformation, the proportion of the two point sets is computed to estimate the position and scale of the bounding box in the current frame accurately and effectively. Experimental results demonstrate that the method is adaptive to scale change and has advantage in illumination variation and color similar target tracking.