Research Of Image-Based Modeling Algorithms

With the development of multimedia and virtual reality, efficient 3D modeling algorithms, which can realize the real-time computation are increasingly needed. 3D model is essential to the object recognition, automatic design, video games, and digitalization of historical relics, reconstruction of historical buildings, and visual scene reconstruction. The reality force the 3D modeling to rethink the complexity and robust of algorithms. In this paper, we mainly discuss the key technology and innovative research of Image-based Modeling System, including Camera Calibration, 3D modeling algorithms and Texture Mapping.In the 1st chapter, we firstly introduce the current application study of 3d modeling. Then we confirm the radix of research in efficient, accurate and robust.In the 2nd chapter, we briefly introduce the basic concept of Image-Based Modeling and Rendering, and the relationships between it and other technologies. After concluding many 3D modeling system, we bring forward the integrated framework of IBMR. It is composed of camera calibration, 3D modeling algorithms and texture mapping. In the mean time, we discuss 3 classic 3d modeling algorithms in detail, including Volumes from Silhouettes, Marching Cube and Dense Stereo Matching.In the 3rd chapter, our research focuses on the method of camera calibration, we mainly describe the DLT system we adopted.In the 4th chapter, we will present some algorithms of polyhedral visual hull reconstruction to make up the deficiency of classic algorithm:(1 Reconstruction algorithm based on ray-tracing, in which all the rays that run through the vertex on silhouette of object will intersect with all the cones correspondent to other silhouette, and we have to calculate the intersection of all such line segments, the endpoints of which are considered to lie on the surface of object and we will get the point cloud on the surface of object. We can see all the calculations in ray tracing are on the rays, and the calculation fall into the intersection between face and line and intersection of line segments, which can be solved by the ordinary geometry operation. Compare to some available method based on the concept of "voxel", our algorithm significantly saves space, the accuracy of approximation is independent of the 3-dimensional resolution, and the speed is also greatly improved especially when the object has large volume.(2) algorithm for topologically generating Visual Hull. It takes the advantage of the topological structure of the object's surface to reconstruct the surface directly instead of computing the extra information of all the voxels inside. Then, an improved SurfaceNet algorithm is adopted to smooth the 3d surface. Compare with the classical method, namely volume carving, thecomplexity of the topological generation algorithm has been reduced to O(n) (where n is the number of vertices on the final surface). What is more, the dependence of the complexity on the photos' number is also reduced. The results of the experiments show that topological generation algorithm is superior to some other traditional visual hull methods such as the octree method, as far as its reconstruction function for practicality is concerned.In the 5th chapter, we study the technology of texture mapping. Based on the basic theory of 2D texture mapping, we discuss the two-part texture mappings, and then provide a feasible mapping algorithmIn the end, all of the works in this dissertation are summed up, and their futures are prospected.