EE 367 Term Project Rendering stereoscopic panoramas from Concentric Mosaics

Project Motivation Cylindrical stereoscopic panoramas give a 360◦ view of the scene around us. They comprise of a pair of images, one for the left eye and other for the right eye, which can be merged and then experienced through a VR viewer, for example the Google Cardboard. One way in which they can be captured is by rotating two slit cameras separated by the Interpupillary distance (IPD) (to stimulate a stereo panorama for human eyes) around an axis which passes through the center of the two cameras. This gives stereoscopic depth cues for all possible viewing directions, but from a fixed viewpoint. If we were to replace each of the slit cameras by conventional cameras that capture multiple vertical ”slits” to from a regular 2D Image, we would be capturing overlapping images of the scene at each angular position. The Concentric Mosaics representation of this data can then be used to theoretically render novel views from any point inside the circle with radius equal to the distance of the camera from the rotation axis. Since one off centered camera, either to the left or the right of the center, would only capture rays in the clockwise or counterclockwise direction respectively, we need a hardware setup that would capture images corresponding to both of these rays for each angular position. This is illustrated for the case of slit cameras in Figure 1 (a) and (b). In [1], authors captured the CM’s by first rotating the camera in clockwise direction and then in the anticlockwise direction. Hence, with only one camera a representation as captured by the two-camera setup was obtained. The advantage of the two-camera setup is to capture both (clockwise and anti-clockwise) rays in one go. More importantly, this project is motivated by the fact that one can generate stereoscopic views from any two points in the circular region separated by the IPD, once such a representation is captured through the two-camera setup as shown in Figure 1 (c).