Research On Two Methods For 3D Reconstruction Of Large Scenes

3D reconstruction technology is to obtain 3D models based on the scene objects’spatial information which is extracted from images. And this kind of techniques is a research hotspot in some fields including computer graphics, computer animation, virtual reality, scientific computing, digital media creation and so on. In the field of 3D reconstruction of large scenes, the vivid virtual scene can be obtained by reconstruction techniques, which make the digital city and digital tourism achievable. Currently, this field has attracted lots of attentions, however, in view of recent studies, there is no unified approach for all cases.Geometric modeling based technique and image-based technique are popular in the field of 3D reconstruction. Geometric modeling based 3D reconstruction technique usually can be used in reconstructing the models by software. The models of conventional constructions always perform well especially for the simple structure buildings. However, this method is non-automatic in the modeling process, and the implementation of the method is inefficient. In view of computer vision, the image-based 3D reconstruction technique can provide automatic 3D reconstruction by extracting the information from images. But, due to the complexity of the generated model, the applications in virtual reality will be limited. Moreover, as the growth of the scene scale, the memory consumption will be growing rapidly.This thesis explored the problems related with the 3D reconstruction of large scenes, compared two existing 3D reconstruction techniques, and proposed several effective solutions to the relevant problems. The main contents of this thesis include the three following aspects.(1) Using geometric modeling based 3D reconstruction techniques, an efficacious method for 3D reconstruction of large scenes is proposed. The thesis discussed the approaches and key techniques of 3D rehabilitation and roaming. The optimization modeling methods are achieved by combining the models with the same properties, using texture mapping and other means.(2) A novel algorithm for 3D reconstruction named MFPMVS (Multiple Features Patch based Multi-view Stereopsis) is proposed. Compared with the classical PMVS (Patch based Multi-view Stereopsis) algorithm, the multiple features of models such as lighting, texture, shadows and concavity are comprehensively considered in the proposed MFPMVS algorithm. And so the models with rich details can be reconstructed. Experiment results show that, the proposed MFPMVS algorithm can obtain more 3D point cloud, and the details of strong concavity model are more delicate. Meanwhile, the loopholes of the reconstruction model with upward-shooting can be significantly reduced, and the edges are more completely. More importantly, the proposed algorithm can rebuild the 3D model of the object more stable and robust, which means the high practicability of the presented algorithm.(3) By using optimized geometric modeling based 3D reconstruction techniques and proposed MFPMVS algorithm, large scenes like Shanxi University campus and Qiao’s Grand Courtyard are reconstructed. And the virtual roaming system have been developed. The result shows that the roaming system running smoothly, and the animation effects are clearly. Which illustrate that for the 3D reconstruction of large scenes, the proposed method is universally applicable, and popularization and application are convenient and easy.