Three Dimensional Vision Based On Rotational Stereo

Compared with traditional stereo vision, rotational stereo takes advantages of lower cost, less calibration and being easy to reconstruct entire three-dimensional objects.This paper does an overall research on rotational stereo including the feature points matching algorithm based on rotational stereo, the 3-D reconstruction algorithms and self-calibration algorithm. Based on the special spatial relationships against rotational stereo, the traditional SIFT algorithm is improved suits with rotational stereo. Results show the improved SIFT algorithm reduces wrong matching rate of feature points.Based on the special spatial relationships among camera coordinate system in the rotational stereo and setting the location of the world coordinate system, two 3D reconstruction algorithms based on rotational stereo are proposed. The first is setting an axis of the world coordinate system superposed with rotating axis of the rotating stage. The second case is setting a plane of the world coordinate system superposed with calibration template. The results of simulation and practical experiments verifying the 3D reconstruction algorithms are acceptable.A rotating self-calibration method against extrinsic parameter matrix is deduced by the relationship among rotation matrix, intrinsic parameter matrix and the coordinate of pole according to the pole imaging principle. That is getting the intrinsic parameter by manual calibration firstly, and then gets the coordinate of the pole through the image information. In the end, we get the direction vector of rotation axis through the relationship among rotation matrix, intrinsic parameter matrix and the coordinate of pole. The translation vector could be got through the relationship among intrinsic parameter, translation vector and rotational matrix. And the extrinsic matrix can be got through translation vector and the direction vector of rotation axis.Rotational stereo only needs one camera to capture object images, which can reduce the cost and the complexity of the operation. Self-calibration method against extrinsic parameter matrix shows that the projection matrix of the camera in any direction can be got if the intrinsic matrix of camera is known. Rotational stereo can reduce the number of calibration and have lower cost. A series of images are easier to be obtained through rotational stereo, which can reconstruct entire three-dimensional objects.