The Design And Implementation Of 3D Outdoor Reconstruction And Navigation System Based On Multi-view Stereopsis

3D panoramic reconstruction is always a key research project in computer vision and computer graphics. Since traditional methods of reconstruction rely on manual modeling that is time-consuming and laborious during the processing, the way how to reconstruct the 3D model quickly and accurately from the real 2D pictures has been a hot topic in this field. In this paper, there is an image-based modeling technique based on multi-view stereopsis which is used to extract visual information from a series of 2D images and to reconstruct a 3D model with the visualization of 3D models in order to preview the 3D model by presenting in 3D interactive way.The main technology of 3D reconstruction is image-based modeling technology that extracts a large amount of image information contained in images and combines relevant parameters of cameras and illumination that are used to figure out the inverse transformation operation of the optical projection and to restore 3D geometric information of the target.When obtain visual information from pictures, there are two methods implemented in this paper that called SIFT algorithm and SURF algorithm which are both applied to extract features from pictures and are compared with each other. Then use the improved neighbor search algorithm based on KD-Tree to match features on every two pictures. Experiments show that this algorithm improve the accuracy and robustness during matching features.When reconstruct the 3D scene information, structure from motion is used in this paper that could recover parameters of cameras and the 3D position of feature points corresponding to the position in 3D space and then optimize results about the accuracy by using the bundle adjustment algorithm. Meanwhile, using camera parameters restored from pictures, it could be provided to work as navigation information for users. It could enhance the convenience and practicability in interactive operation.In addition, CMVS/PMVS algorithm is adopted to reconstruct dense models that cluster the sparse model and carry out pairing, expanding and filtering repeatedly in order to get dense 3D models. Results show that dense models generated without any prior knowledge are accurate and valuable.Finally, there is a 3D panoramic reconstruction and navigation system based on multi-view stereopsis designed and implemented in this paper. It realizes the method that detect and match feature points, structure from motion and CMVS/PMVS algorithm that are mentioned in this paper.What's more, it provides a visualize 3D scene with navigation and preview functions by visual processing.