The Research Of Stereo Matching Algorithms In Different Media

Binocular stereo vision technology is aimed at rebuilding the 3D scene. It has the advantages of non-contact, high accuracy, large amount of information and so on. And it has been widely used in various fields such as robot navigation, aviation mapping and sea exploration. In stereo vision, the stereo matching is the most important and difficult step, matching accuracy and speed will directly affect the application of stereo vision. This paper conducts systematic researches on binocular stereo vision and makes improvement to the existing stereo matching algorithm both in the air and water. The main works are as follows:(1) The paper starts with the cameral projecting model, and introduces the linear model, nonlinear model and refraction model of underwater. At the same time, it introduces transitions of the coordinates. Meanwhile, the parallel model in binocular stereo vision is established and the relationship between the parallax and depth is given. The basic principle, steps, constraints of disparity and the classification of stereo matching algorithms are introduced as well.(2) Due to the limitations of single matching cost algorithm and the bad adaptive of fixed parameter combined method, a new combined cost based on adaptive parameters is presented. Firstly, use SAD and Census algorithm to calculate the matching cost for each pixel separately. Secondly, calculate the mean and standard deviation of cost for each pixel within the disparity range, and then the weight is obtained. Finally, according to some certain relationship, combine the two algorithms together with the corresponding weight, and then the new cost function is obtained. Experimental results show that the proposed method combines the advantages of SAD and Census algorithm, so the matching error rate reduces about 12%. And the cost function can choose the optimal weight adaptively, which shows good adaptability for the weak texture, repetition texture and changing illumination situations. The paper laid a solid foundation for the global and local matching algorithm.(3) In terms of underwater non-parallel binocular image matching cannot be parallel correction, and no longer satisfy the epipolar constraint in the air, a kind of epipolar constraint model that suitable for underwater is presented. According to the differences of relative position between the camera and waterproof cover, we divide the underwater imaging ways of binocular vision into parallel imaging, common reflection surface imaging and independent reflection surface imaging. And we deduce the epipolar constraint model of the latter two. Firstly, we calibrate the camera. The scale invariant feature transform(SIFT) algorithm can help to match two images. And corresponding curve of the feature points on reference image is derived. According to whether the corresponding feature is on the curve, we can eliminate false matching points and improve matching accuracy. The experimental results show that the underwater curve constraint model can effectively eliminate false matching points, and reduce the error matching rate.