Research On Stereo Matching Algoritnm For Panoramic Stereo Vision

Panoramic stereo sphere vision model is constructed by using four fish eye lens cameras with perpendicular optical axis,and global image can be observed.Fish-eye lens has large FOV and distortion,especially more serious in the edge region.It is difficult for image feature matching in such a large field of view and orthogonal axis stereo vision system.At present,fish-eye image matching usually corrects the distorted image into perspective projection image,then finishes the match.This method usually results in the loss of image information,may increase the mismatch.Therefore,it directly matches fish-eye image without distortion in this paper,The main research contents include:(1)In this paper,the feature descriptors of the combination of MSCR and ASIFT are used to study the stereo matching of fish-eye image.The stable regions detected by MSCR are more regular and the global effect is clear and bright,which is more conducive to feature area matching.By comparing the performance of MSER and MSCR in stereo matching without distorting corrected fish-eye images,it is found that the matching performance of MSCR is better,the matching points are more,and the false matching rate is lower.ASIFT not only inherits the invariance of the scale,rotation and translation of SIFT,but also can realize complete affine invariance.The performance of ASIFT is evaluated and analyzed objectively,and it is found that it can detect much more features than other algorithms.(2)The strategy for realizing stereo matching is:First,spatial target is collected to get stereo image pair.One image is defined as standard image,and the other is matching image.Then a point A is selected in standard image and area DR,which contains corresponding point in matching image,the stable and invariant description region R(A)is built with A as the center,ellipse is synthesized in R(A).Meanwhile,the DR is partitioned according to the size of R(A);Then ASIFT is used to describe R(A)and the block,so the one dimensional feature description is obtained.Finally,matching results are obtained by comparing the similarity of histogram.(3)In this paper,the feature extraction and stereo matching of images are carried out on the MATLAB R2013 b platform.It is proved by experiments that the algorithm used in this paper can detect more matching points,the matching quality is better,the mismatch rate is lower,and can be applied to stereo matching without distorting corrected fish-eye images,provides effective way for the stereo matching with super distortion and visual angle,and lays a foundation for subsequent 3D reconstruction.