2D-to-3D photo rendering for 3D displays

We describe a computationally fast and effective approach to 2D-3D conversion of an image pair for the threedimensional rendering on stereoscopic displays of scenes including a ground plane. The stereo disparities of all the other scene elements (background, foreground objects) are computed after statistical segmentation and geometric localization of the ground plane. Geometric estimation includes camera self-calibration from epipolar geometry, and an original algorithm for the recovery of 3D visual parameters from the properties of planar homologies. Experimental results on real images show that, notwithstanding the simple “theatrical” model employed for the scene, the disparity maps generated with our approach are accurate enough to provide users with a stunning 3D impression of the displayed scene, and fast enough to be extended to video sequences.