Abstract
A technique is presented for modeling 3-D shape of rigid as well as non-rigid objects employing multiple video images. Several feature points are initially specified on the object interested. The image coordinates of the feature points on the video images are collected into a single matrix called an extended measurement matrix, to which singular value decomposition is applied to obtain the 3-D locations of the feature points. These recovered feature points provide us a 3-D model of the object. Advantages of the presented technique over others include that neither camera calibration nor registration of recovered feature points is required and that it can be applied to any object only if feature points can be defined on it. Experimental results are shown on some objects including human motions.
