Scene Rendering Method To A ect Motion

Techniques to generate novel scenes have been proposed in computer vision research. In addition, there have been a lot of 3D models written in VRML for viewing on the WWW. However, the depth of 3D images cannot easily be sensed without special hardware such as HMDs and without special equipment, e.g., shutter and poralized glasses. In this paper, a method is proposed that uses motion parallax as a depth cue to let users perceive the 3D structures of models rendered into images. In this method, the user's head is tracked and the motion parameters are estimated by using a computer vision technique. Then, the 3D object is rendered from the estimated eye position as a new viewpoint. The system has been build on an SGI O2 workstation with its mounted camera used for head tracking. The experimental results are very good; users feel as if the 3D objects are placed in front of the screen of the monitor.