Abstract: We present in this paper a method to estimate movementin digital video sequences from a 3D representationmodel based on polynomial transforms. This transformallows to obtain representations of the video sequenceat multiple spatiotemporal resolutions. Our approachanalyzes the video sequence locally by means ofspatiotemporal windows. It allows to recover the trueflow in regions free of the aperture problem as well asgiven an estimate of the normal component of motion fororiented patterns detected within each window.Different window sizes in space and time are allowedfor an efficient analysis of all types of motion.!24
展开▼