Adaptive parallax control for multi-view stereo panoramas

摘要

We describe techniques for stereo panoramic image capture and rendering that are part of a personal panoramic virtual environment system. We examine the use of stereo shutter-glasses and several recently developed autostereoscopic (AS) displays to improve the sense of immersion. The stereo panorama pair is created by stitching strips that are sampled from images captured with swing camera panoramic imaging system. We apply Peleg's disparity adjustment algorithm to the generated stereo panorama to achieve large disparity (horizontal parallax) of far away scenes and smaller disparity of closer scenes for stereo perception. Unfortunately, vertical parallax effects in the stereo panorama still occur, causing display artifacts and problems in human stereo fusion. To overcome these problems, we first present a general image capture model, specify geometrical parameters, and describe the panorama generating process. We then describe an efficient stitching algorithm that corrects dynamic exposure variation and removes moving objects without manual selection of ground-truth images. We present expressions for the horizontal and vertical parallax, describe parallax measurement techniques, and develop an adaptive vertical and horizontal parallax control algorithm for rendering in different viewing directions. We present a simple subjective test of stereo panoramas rendered on AS and other stereo displays, and discuss the relative quality of each.