Based Multi-view Three-dimensional Reconstruction Study

With the bloom of film and television animation industries and the vast emergence of visual publications, 3D reconstruction, one of the key techniques in this area, have received extensive attention and owns a promising future of development. However, traditional 3D reconstruction technique faces several roadblocks as follows: difficulty in environment building, the comparatively higher cost, lower flexibility, and insufficiency to adapt dynamic modeling object together with the requirement for popularization. Image-based 3D reconstruction technique provides a convenient and practical method for 3D modeling and multi-viewpoint 3D reconstruction is one of the hot topics in this field. In this paper the multi-view 3D reconstruction method is explored from the perspective of computer vision, propose and realize the algorithm of geometric- constraint-based spatial point reconstruction is discussed in depth, which effectively improves the precision, reduces the requirements of environment, satisfies the demand of dynamic reconstructing process, and lowers the cost of dynamic 3D modeling. As a result, this method is helpful to stimulate the universal application of 3D reconstruction technique in the related entertainment industries and visual publications.In this paper after analyzing the state-of-art of 3D reconstruction method in animation industries and visual publications, I propose the research route that uses the common digital camera as image capturing tool and select properly multi-view reconstruction algorithms to guarantee precision and at the same time reduce loss of image features. My research is centered around multi-viewpoint-based object's 3D feature extraction and 3D reconstruction. For the purpose on utilizing common camera for precise 3D reconstruction, we firstly regulate image distortion with the non-linear modeling of camera, which efficiently eliminates the reconstruction errors caused by radial distortion of camera. Then we improve the algorithm of Hough Transform for circle detecting in order to extract features, with which we nearly achieve the degree of pixel precision and enhance effectiveness of the extraction process. Next we apply the two-step method of non-linear model for camera calibration, with which we are able to properly regulate distortion of collected images and to further upgrade accuracy of feature extraction. Finally we forward and realize an algorithm for 3D spatial point reconstruction using Euclidean Distance as constraint condition. It is proved that this algorithm can reach rapid convergence in Euclidean Space while guaranteeing comparatively high precision.To validate the efficiency of above approach, two separate cases of 3D reconstruction to polyhedron and curved surface are presented in this paper. The results prove the effectiveness of improved multi-view 3D reconstruction method in satisfying the needs of flexible and convenient object reconstruction with common camera in multiple situations, with the proper precision provided. Furthermore, application in the development of a virtual interactive system-"History of Computer (Magic Book)"brings preferable results in 3D reconstruction and virtual interaction.