Dense disparity estimation from stereo images

Binocular stereovision is based on the process of obtaining the depth information from a pair of left and right views of a scene. In this paper, we describe a new approach for comput-ing a dense disparity field based on a convex set theoretic for-mulation. The stereo matching problem is solved through the minimization of an objective function under various convex con-straints arising from prior knowledge. In order to preserve the discontinuities in the disparity field while getting a stable solu-tion, we consider different appropriate regularization constraints. The resulting multi-constrained optimization problem is solved via an efficient algorithm which was recently introduced in the convex optimization literature. Experimental results on standard data sets demonstrate the validity of the proposed approach. 1