The goal of this thesis is to research on dense depth map acquisition by passive methods, i.e., stereo matching methods. Given two images of the same scene captured in two different views, the depth map can be obtained by finding the corresponding point in the target image for each pixel of the reference image. The depth map is used in a variety of applications, including mobile robotics navigation, 3D entertainment, aerial survey, 3D machine vision for industrial automation and so on.
The stereo matching methods can be generally divided into two branches as, global and local methods. This thesis work is focused on local stereo matching methods, which consider correlations between intensity patterns in support windows. That is, the raw matching costs of all pixels within a support window are aggregated to the center pixel at each disparity; then an optimal disparity that gives a minimum aggregated cost is selected through an efficient local optimization process. Until now, we have studied the state-of-the-art local methods and proposed a fast and accurate method, whose effectiveness is demonstrated by the evaluation on the standard benchmark.