Research On Stereo Matching Algorithm Based On The Feature Of Edges

Binocular stereo matching is the process of obtaining depth information from computing offset of the pair of images obtained from two different viewpoints of a scene, through image matching and triangulation measuring principle. Stereo matching is the critical problem in stereo vision and core in 3-D reconstruction. Feature-based stereo matching extracts kinds of features of images as matching primitives, so it does not depend on illumination directly, which can have strong anti-interference performance. It is also quick to compute, can meet the real-time request.In the article, firstly, we instruct some feature extraction methods used in feature-based stereo matching algorithms, and select the zero-crossings acquired from convolution of LoG filter as matching primitives. In one side, we can define the size of LoG filter using the psychophysiology data obtained from human visual system, so it can simulate human vision. In the other, we use the particular attributes of zero-crossings as constraint in coarse matching, which can improve the matching speed.Secondly, we analysis the main constraints and similarity criteria used in stereo matching. We proposed figure continuity constraints by applying disparity smoothness constraint in the edges extracted by our algorithm, which can eliminate the error matching; and the similarity measure of Rank transform, which is non-parametric transform, can improve the reliability of matching.Thirdly, we proposed the stereo matching based on the feature of edges. The algorithm used the particular attributes of zero-crossings, sign and direction, in the coarse matching, and figure continuity constraints to eliminate the error matching. After the coarse matching we use the adaptive stereo matching algorithm based on edge detection in the zero-crossings which have no or more than one correspondences.