Investigation On Stereo Matching Techniques In Micro Stereo Vision System

Stereovision studies the recovery of three-dimension information of an object through its perspective projections, and has been widely applied in industrial manufacturing process for various tacks that require machines to be able see. Artificial stereovision systems imitate the human vision system in the sense that two or more images are fused to produce depth information; but they still can't reach the performances of the human vision. However human vision is limited in size and working environment.Micro stereovision is a branch of stereovision that investigates artificial vision problems for small size objects. The recovery of depth information includes several steps such as system calibration, image processing, stereo matching; etc. Stereo matching is the most important and the most difficult step. One of the main problems in stereo matching is how to deal with occlusions that are unavoidable for every case. Stereo correspondence is also affected by image noise and lightning conditions. Despite major progress already made in stereo matching techniques, there is no a standard and robust stereo matching algorithm that would be applied for every presented scene. In this thesis, we analyze the structure and applications of artificial vision systems. We also propose a stereo matching algorithm for micro stereo images that uses dissimilarity measure between matching pixel candidates. Before the matching process, a vector medium filter is used to denoise the stereo images, and the background is pruned out to remain with the pattern to be analyzed. The proposed algorithm is tested on real micro stereo images and the results are displayed in a three-dimension model.