Detection_and_classification_of_motion_boundaries英文文献资料.pdfVIP

Detection and classification of motion boundaries Richard Mann Allan D. Jepson School of Computer Science Dept. of Computer Science University of Waterloo Univerity of Toronto Waterloo, ON N2L 3G1 CANADA Toronto, ON M5S 3H5 CANADA Abstract We segment the trajectory of a moving object into piece- wise smooth motion intervals separated by motion bound- aries. Motion boundaries are classified into various types, including starts, stops, pauses, and discontinuous changes of motion due to force impulses. We localize and classify mo- tion boundaries by fitting a mixture of two polynomials near the boundary. Given a classification of motion boundaries, we use naive physical rules to infer a set of changing con- tact relationships which explain the observed motion. We show segmentation and classification results for several im- age sequences of a basketball undergoing gravitational and nongravitational motion. Introduction Given the trajectory of a moving object, what physically Figure 1: A composite of the tracking results for a sequence meaningful aspects of the motion can be recovered? Here where a subject throws a basketball. we segment the trajectory of a moving object into piece- wise smooth motion intervals separated by motion bound- aries. We consider the motion of a single passive object, termine the allowable motion at such boundaries. Here we such as a basketball, undergoing gravitational and nongravi- present a simpli