Efficient Stereo Matching Approach Using Pixel Selection Patterns in Stereo Vision

Stereo vision finds a wide range of applications in areas of computer and machines, such as providing three dimensional perceptions, imitating humans. The crucial correspondence problem faced by most visual systems can be resolved by Stereo matching algorithms to a great extent. Broadly, the matching algorithms are categorized as area based and feature based. Area Based technique involves correlation based matching which is relatively fast and provides a dense disparity map. In this paper, some significant techniques involved in area based matching are discussed. This paper describes the computational optimization strategy, which is based on a very effective pixel selection scheme. Normalized Cross Correlation (NCC) has the advantages of accuracy, strong robustness, but speed is its weakness, so online application is restricted. This method uses novel pixel selection schemes where disparity map produced using NCC is modified in such a way that provides accurate and computationally efficient results. Finally, we provide experimental results obtained by the proposed methods on variable window sizes; we compare these data with those obtained using a well-known, fast, area-based algorithm SAD. The results show that the proposed method is computationally faster than the conventional NCC scheme.