Study on the optimization algorithm of polar line correction based on feature extraction and Gaussian filtering

Aiming at various common problems in the process of calculating polar line correction in the matching direction of binocular stereo vision, this paper proposes a polar line correction algorithm based on gray-scale change threshold feature extraction and Gaussian filtering to filter out noise points. In binocular stereo vision, the feature point matching problem of stereo images has always been its core. In this paper, after extracting the feature points with the reference of Gluckman's method and Francesco's method, we use the improved normalized eight-point algorithm to solve the base matrix and reject the mis-matching points with RANSAC algorithm, so as to avoid the non-parallelism between the polar lines caused by the external influence such as assembly, i.e., the vertical parallax of the corresponding feature points during binocular measurement. The image is then transformed by projection, so that the images correspond to the same scan line, thus satisfying the scanning characteristics.