Image Based Modeling Technique Research

Nowadays, image based modeling and rendering (IBMR) is a quite active research direction in computer graphics field, which can reconstruct the 3D scene directly using images and conquer the shortage of traditional geometry-based graphics. There is a great challenge to work out a robust, precise, stable, and applicable image based modeling and rendering algorithm, which is both theoretically and practically valuable.In the thesis, we study at IBMR technique. The main work can be summarized as follows:1. Lines Extracting and Matching Technique: Straight line, which is widely used in computer vision, is very important middle-level symbol in image analysis. There are many approaches to extracting and matching straight lines, but they are all time-consuming, which limit their uses in the Real-time system. Aimed at the Real-time application, such as robot teleoperation system, a method for robust and fast extracting and matching line segments based on support region of lines is proposed in the paper, The validity of this approach is indicated by experimental results.2. The Regular Objects Modeling Technique: The paper enhances the precision of modeling, reduces the manual work and realizes the fast modeling algorithm for the regular objects using the extracting lines of two images and the geometric limited condition based on objects. The algorithm realizes the automatization of modeling in a certain extent.3. Evaluating the Visual Hull Technique: The paper mades deep study on the principle of Visual Hull and a new approach, that is multiple images are used to generate Visual Hull models, is proposed in the paper. Realistic models of 3D scenes can be neatly obtained with our method. Compared with other approaches, the method is provided with larger advantages, aiming at modeling from the satellite images being of single background. We take a test for the different complexity's objects and practice shows that the results are correct and reliable so as to meet the requirement s of virtual reality applications.4. Triangulation Technique Based the 3D Points Cloud: The paper improves and realizes a new approach of directly triangulation of 3D points cloud aim at the points models of 3D objects acquired from Visual Hull algorithm. We carve up the 3D points using cubic mesh, so as to reduce the searching area, improve the correctness and confirm the quality of the triangular mesh. The method is proved to be highly exact and efficient.Based on the above research fruits, the paper realizes a simple modeling system (Stereo Vision) based on the matching lines and an integrated modeling system (VH) based on Visual Hull, which offer a credible experimentation platform for the farther research.