Research On The Technology Of Walking Through Virtual Real Scene Based On Single Image

Image-based modeling and rendering is a fascinating technique in recent years, and now it is also the essential technique in the field of walking through virtual scene. It is the most remarkable that IBMR can overcome the deficiency of traditional geometry-based modeling and rendering technique. And IBMR can realize photorealistic modeling and walking through in the computer with common computational capability.In the field of image-based modeling which is based on the theory and method of computer vision and image processing, reconstructing from a single image is an important topic of research. This thesis addresses the issues of calibrating from a single image, reconstructing 3D models of the scene, rectifying texture from the photograph and mapping rectified texture.First, in the scheme of calibrating the single image, three vanishing points of orthogonal directions can be determined by means of human-computer interaction, and to improve the Guillou algorithm, calculation of the position of principal point is introduced. By which, we can not only obtain the focal length of camera by using the simple geometrical relation and the rotation matrix from the world coordinate system to the camera coordinate system, but also improve the precision of the calibration result.Secondly, in the scheme of reconstructing 3D models of the scene, this thesis parameterizes the cuboid structure of the scene and uses the focal length, which is calculated in the calibration phase, to mend the result of this calculation. So we can obtain the relative depth of those vertexes, three parameters of the cuboids structure and the translation vector of the world coordinate system. The simple 3D models of the scene are reconstructed under a global scale factor.Then, in the scheme of rectifying and mapping texture, this thesis presents an improved method which is based on the theory of stratified reconstruction of 3D vision and combines the features of texture mapping function of OpenGL. Texture images which are extracted from the photograph are changed into square images with 2~n side length. Then the texture images are mapped into the surfaces of simple 3D models. So we can obtain photorealistic 3D models.