Research On Feature Extraction And Matching Method Of Spherical Stereo Vision

Compared with ordinary lens, fish-eye lens has larger field of view, and even direct access to the image information of the hemisphere domain. In the field of stereoscopic vision, using fish-eye lens to get a panoramic image can reduce the number of lens and image acquisition modules, simplifying the system, increase speed, and reduce costs. So it is a new research project in the panoramic field of vision. However, images taken with a fisheye lens will produce a certain amount of distortion, and the more closer to the edge of image the more larger to the distortion. Therefore, it is difficult to match the feature points of the image in the stereo vision system with the orthogonal or the angle of the optical axis, which directly affects the application effect of the stereo vision system. Process of fish-eye image matching mainly includes several steps, which are feature extraction, descriptions, matching and removal of mismatch. Aiming at the characteristics of fish-eye lens, this paper carry out research on image matching technology. Main work includes:1, traditional fisheye image matching method is matching feature after image distortion correction, resulting in loss, fracture texture, and arch distortion of image information. Aiming at the shortage of traditional methods, this paper presents an improved method, namely direct on the distortion of fisheye image feature matching and can be directly 3D reconstruction, or is the spherical image registration and fusion after correction to image mosaic.2, for distortion of fish-eye image, an image matching algorithm with affine invariant must be used to the feature matching. In feature matching of distorted image, first extract the MSCR region of the original image, followed by the introduction of CS-LBP operator on the MSCR regional image feature extraction, and then divide the image into several sub-areas, by connecting the CS-LBP histogram of these sub-areas to get the characteristics vectors.Chi-square distance algorithm based on feature vectors used to only match features, finally,the region is adjusted and marked for attachment of the ellipse.3, illumination, complex scenes degree, shooting error and other factors will cause number of feature regions and matching pairs, leading the increase of overlap by characteristic areas and mismatching rate. This paper shows advantages of image matching algorithm by comparing the mismatch rate of feature detection and description., which mixed programming of MATLAB and C + +.