Research On Method Of Catadioptric Panoramic Stereo Imaging And Depth Estimation

Stereo vision based on human visual principium explores the method ,which estimate 3D information from 2D images, in order to understand 3D world from planar images. The field of view in traditional optical imaging systems is usually small, thus the depth information obtained by using stereo vision techniques is limited, which is inconvenient in many applications. As the techniques of navigation of mobile robots, autonomous surveillance, virtual reality developed in recent years, stereo perception and reconstruction of omni-directional scenes with large field of view by using the techniques of stereo vision and panoramic imaging attracts many researchers. Catadioptic panoramic stereo vision can obtain panoramic stereo image of omni-directional scenes, estimate depth information, simplify algorithms and speed up processes.A system with single camera and two paraboloidal catadioptic mirrors for estimating depth information of panoramic stereo images is designed. The following problems are solved:(1) A panoramic stereo image-creation system is designed, which uses paraboloidal catadioptic mirror. The system is set up of a single camera and two paraboloidal catadioptic mirrors, which are coaxial in vertical direction. This design makes the images by the two paraboloid catadioptic mirrors are concentric circles. Thus objects in the constrained fields will get a couple of pixels, through which depth information can be estimated.(2) A formula is figured out to calculate depth information according to the design of the panoramic stereo image-creation system. The parameters of the structure of the system in the formula are used as independent variables to analyze system errors. Results of this analysis are used to improve the system structure.(3) Making the original omni-directional stereo image into panoramic stereo images by projective transformation in order to make the use of the method of corresponding match in existence, improve precision of the corresponding matching and simplify the algorithm of depth information estimation.(4) Two kind of algorithms: SAD and DP are implemented, which are used to match corresponding points. Omni-directional stereo image is obtained in virtual and practical scenes separately using the design in this paper. Depth information of the scenes is estimated by these two corresponding match algorithms, and the dense disparity image is obtained. This prove the system and the algorithms are available.