Occlusions and Binocular Stereo

Abstract

Binocular stereo is the process of obtaining depth information from a pair of left and right cameras. In the past occlusions have been regions where stereo algorithms have failed. We show that, on the contrary, they can help stereo computation by providing cues for depth discontinuities. We describe a theory for stereo based on the Bayesian approach. We suggest that a disparity discontinuity in one eye's coordinate system always corresponds to an occluded region in the other eye thus leading to an occlusion constraint or monotonicity constraint . The constraint restricts the space of possible disparity values, simplifying the computations, and gives a possible explanation for a variety of optical illusions. Using dynamic programming we have been able to find the optimal solution to our system and the experimental results support the model.