Abstract
Three-dimensional shape recovery is one of the major topics in computer vision. It has application to object modeling in virtual reality or in electronic museums. Human motions are recovered by the technique to produce their 3-D models in a motion capture system, which are employed in 3-D video games, sports form analysis and rehabilitation training evaluation. Existent shape recovery techniques such as stereo vision recover entire shape of an object by combining locally recovered shapes. This will inevitably include connection errors. In this paper, a novel technique is proposed for recovering an entire shape of an object by employing virtual see-through cameras surrounding an object interested. Camera calibration of the proposed technique is done based on images: This simplifies the use of the technique to a large extent. An example of 3-D human motion modeling is shown, which gives all the views of the motion.
