Enhancing the volumetric approach to stereo matching

We propose techniques to enhance the volumetric approach to stereo matching, with which to obtain dense disparity and detect occluded zones. The volumetric approach works on the row × column × disparity space initializing the voxels with a measure of the similarity between the stereo pair that they represent (similarity phase). In this phase, we propose a function L₀ that provides a better initial estimation for disparity. Then, these values are refined through an iterative process which inhibits all but one voxels placed along the same line of sight (competition phase). In this phase, we propose to use a dynamical programming technique for faster converging to the final solution and producing smoother maps. We substantially reduce the proportionality constant of the original algorithm complexity, enhancing time without strongly influencing the results