| Binocular vision stereo matching technology is an important branch of computervision. It is widely used in areas such as aerial mapping, robot navigation forplanetary exploration, three-dimensional reconstruction and so on. The most critical anddifficult technology of binocular vision is the research and selection ofstereo matching algorithm since the efficiency and accuracy of matching will directlyaffect the application of stereo vision technology.This paper improves the accuracy of SAD algorithm and apply the algorithm offeature matching into underwater environment based on the research of the correlationstereo matching algorithm. Main work and study are as follows:(1) In this paper we introduce the transitions of different coordinates and parallelmodel in binocular stereo vision based on the camera projection model. Meanwhile,rationales and related theories of stereo matching are outlined. The features anddifferences between basic steps of stereo matching, local matching, and global matchingare summarized in this paper. Moreover, we also introduces the evaluation criteria ofstereo matching results.(2) There exist some problems for SAD algorithm. For example, the match windowis greatly affected by the central pixel and the results are easily influenced by the noise. Tosolve these problems, the paper proposes a Rank transform of stereo matching based onweighting window function. Firstly, the nonparametric Rank transform are performed onboth the left and right gray images to suppress noise. To retain as much the detailedinformation of the scene as possible, we apply the weighting window function into Ranktransform. Secondly, in order to eliminate the mismatching points, we constrain thechromatic aberration index to match again in color space and eventually get thedense disparity map.Simulation results demonstrate that algorithm in this paper hasbetter ability to restrain noise and enhances accuracy compared with SAD matchingalgorithm.(3) In the light of underwater binocular image matching cannot satisfy the epipolarconstraint of air, and the mismatching rate of underwater image processed by the SIFT algorithm is high, this paper puts forward an underwater feature matching algorithm basedon curve constraints. First, binocular camera should be calibrated, to obtain some relevantparameters, the reference image and the matched image. Then the SIFT feature matchingalgorithm can match these two images, simultaneously, the feature points can be extractedfrom the reference image to deduce the corresponding curve on the matched image. At last,the curve is used as a constraint to determine whether the corresponding feature is on it,thus mismatching points will be excluded to achieve a higher accuracy. The results of thetest show that this algorithm is superior to SIFT algorithm and can exclude mismatchingpoints effectively. The matching accuracy can be increased by nearly12%for thematching method in this paper. The problem of SIFT algorithm’s high rate of mismatchingfor underwater binocular stereo matching is solved. |
