Research On Rectification Algorithm Of Stereo Image Pairs

As an important branch in the compute vision, Stereo vision is all the time one of key and hot points. Stereo matching is a difficulty in the stereo vision field and is the bottleneck of stereo vision at present. Epipolar constraint is one of constraints commonly used in the stereo matching algorithm, so epipolar rectification becomes one of the commonly technique accelerating stereo matching. The searching range of stereo matching algorithm is reduced from 2-D image plane to 1-D scan-lines by epipolar rectification, and thus greatly improves the speed and accuracy of searching.This paper makes a study on the rectification of stereo images pairs. The following works have been done:(1) In the process of researching the binocular rectification, the classical algorithm proposed by Hartley is introduced and the experiments are carried out separately according to two different methods of point-selection, and it turns out that under the condition of accurate matching points and fundamental matrix, the algorithm has a good performance. At the same time, by using the principle of layered rectification algorithm, an image rectification algorithm based on immune monoclonal strategy is proposed, and it is compared with algorithm without using immune monoclonal strategy and algorithm proposed by Hartley, which shows that the algorithm proposed in this paper has marked improvement in terms of vertical difference and image distortion.(2) In the process of researching the trinocular rectification, by extending the SIFT and RANSAC algorithm which solves the matching points and fundamental matrix between the image pairs during binocular rectification, the method for three images is proposed. By putting this method into use in the trinocular rectification, it shows that this method is simple, effective and robust.(3) In this paper, the trinocular rectification algorithm is divided into two different thoughts, from which the representative algorithm is selected for research. The first rectification method presented by Heinrichs is based on accurate solution of fundamental matrix, and the effect of experiments is not ideal, but the second method proposed by Sun gets the satisfactory results. Through the experiments on algorithm related to Sun, a thought for improving the rectification results is proposed and validated by experiments, and the experiments show that when the epipolar lines on the original images pairs are approximately parallel, the practice that the shearing transformations used in the rectification algorithm presented by Heinrichs are added to the algorithm proposed by Sun can further improve the precision of rectification and will not increase the image distortion.