The Research Of Stereo Matching Algorithms For The Binocular Vision Based On Fisheye Lens

Binocular vision technique is used to get 3D information based on image pair. Since this technique can get a wealth of information without touch, it has been widely applied in industrial detection, robot navigation and ocean exploration, etc. Compared with traditional binocular vision, fisheye binocular vision system has advantage in getting image information of wide field view in real time. Stereo matching is a key step for binocular vision. And, the practicability and the value of binocular vision technology are determined by matching accuracy and speed. It is difficult to get accurate matching results of fish-eye images due to the problem of high distortion of image. This paper conducts systematic researches on fisheye binocular vision system. Then a fish-eye image matching algorithm is proposed. The main work and study are shown as follows:(1) This paper starts with introducing the background and significance of stereo matching. Second, introduce the fisheye binocular stereo vision research status and existing problems. Finally, the conventional image matching framework of stereo matching is analyzed and the content of this thesis is determined.(2) Due to the problem of traditional polar constraints do not apply to the fisheye image. This paper proposes a fish-eye image matching algorithm based on polar geometry. Firstly, fish-eye imaging model is introduced and the parallel fisheye binocular vision system is established. Then fisheye images of 3D and 2D polar equation are derived according to the fisheye binocular vision model. Finally, this thesis use the dynamic programming algorithm to realize fisheye image of dense matching according proposed polar equation. The experimental results show that our algorithm can get better dense disparity map.(3) The high distortion of fisheye image makes the correspondence between two matching windows can not be guaranteed, it brings a lot of difficulty to stereo matching of fisheye images. This paper proposes a new method to determine the support neighborhood, which can be used to calculate the matching cost. Firstly, a spatial neighborhood is selected for matching pixel based on imaging feature of fisheye binocular stereo vision system. And the center of spatial neighborhood is corresponding spatial point. Then, expression of the spatial neighborhood is calculated on each image under different disparity. Finally, matching cost is computed by using the support neighborhood to realize stereo matching of fisheye image. The experimental results show that the method proposed in this paper can obtain almost of all pixels' support neighborhood to compute accurate matching cost.