The purpose of this project is to display photo-realistic three-dimensional images using off the shelf cameras, and with minimal camera calibration. Three images are captured, and the position, orientation, and calibration of the cameras is computed from corresponding point matches. The geometry is then used to automatically determine matching edges between the three images. These matching edges are interpolated to generate images from intermediate viewpoints.
The position of the viewers head could be tracked using a video camera mounted on the display, and as the viewer moves his head, the display would be updated to show the corresponding viewpoint.
One of the difficulties of using an edge based representation is that edge matches do not directly define how near and far surfaces interact in the presence of occlusion. An occlusion occurs whenever an object nearer the viewer obscures a more distant object. The novel contribution of this work is the detection of occluding boundaries in the images, and proposes a method for utilising this information to correctly render images with occlusion.
With occlusion detection