New adaptive tracking algorithm of non-rigid objects

We describe a contour tracking method based on the Hausdorff distance to track non-rigid objects moving in two-dimensional images. The method operates by extracting two-dimensional templates of the non-rigid objects which vary from one frame to the next, from a sequence of images. The main idea of this method is to decompose the variation of the non-rigid object contours in space into two components: two-dimensional shape changes and two-dimensional motions. The two-dimensional motion components are factored out by the match of the next frame image with the templates which are transformed by a group of transformations such as translation, scaling, and rotation. The two-dimensional shape change components are achieved by setting a proper threshold to allow some biases, so the template adaptability is improved. The two major assumptions underlying the method are: the two-dimensional motions are in the scope of the images; the two-dimensional shape of the objects will change slowly from one frame to the next. Combined with the pyramid algorithm and zero-crossing detection algorithm, experiments conducted on human contour tracking are reported