Search

Based on OpenGL, this paper presented a local 3D scene reconstruction from the frame buffers (including color buffer and depth buffer) after 3D scene rendered. According to projection principles and depth buffer characteristics, it discussed an inverse prooection processing and obtained theoretically a 3D vertex reconstruction method. As a result, it transformed a l valid pixels within frame buffers and generated a 3D vertex set which was a point cloudy model in the camera space in the conducted experiments. Obviously, this cloudy model was a kind of a sample of the original 3D scene in current camera system and named as virtual vision model(VVM). Secondly, it analyzed the point-wise precision of VVM detailed. The results of experiments show that VVM appears very high relative accuracy to the original 3D scene. Above all, the reconstruction method appears natural multi-resolution feature with respect to the resolution of frame buffers and supports scene simplification and streaming transmission. On the other hand, this method is capable of abstracting VVM real-time from the common 3D software easily without any details of 3D scene known previously, which provides a new approach acquiring 3D data efficiently. [ABSTRACT FROM AUTHOR]

The virtual 3D space is a digital one of real world and the human stereo visional space is a 3D reconstruction of the real world or virtual scene through human visional system. Traditionally, human views the environment around himself directly and the geometric relationship between the human stereo visional space and the real world exists undoubtedly. How about the relations of stereo visional space to the virtual 3D space are？ According to the principle of binocular disparity, a geometric model of stereo vision is established for the virtual 3D scene and the corresponding visional 3D model is constructed too. And the screen disparity and retina disparity are analyzed with respect to an arbitrary spatial point. As a result, the geometric mapping between the point of virtual space and the one in visional space is established with matrix algebra, indicating that the 3D visional model can be mapped to a virtual 3D model unambiguously and the metric capability of human visional system is intrinsic. The main contribution of this paper is that a wholesome visional 3D model is proposed, which lays down a quantitative analysis foundation for the traditional stereo vision. The results are useful theory explorations for the stereo presence, the virtual interaction and the metric applications. [ABSTRACT FROM AUTHOR]