Font Size: a A A

Research On Related Technologies For 3D Reconstruction From Uncalibrated Image Sequences

Posted on:2010-03-31Degree:MasterType:Thesis
Country:ChinaCandidate:Q ZhouFull Text:PDF
GTID:2178360278959325Subject:Traffic Information Engineering & Control
Abstract/Summary:PDF Full Text Request
The technique of image-based modeling is a new technique combined computer vision with computer graphics. This technique can reconstruct 3D model from image sequences. The early approaches have to achieve the camera calibration firstly and then can complete the residual work, therefore it can deal with the reconstruction in still or the known conditions, then it has less practicability and it's hard to promote the application. The technique of 3D reconstruction from uncalibrated image sequences doesn't need to measure the camera parameters beforehand; the 3D models of objects and scenes can be reconstructed from several digital images which captured by hand-held camera moving freely. This method has broad prospective in many fields like reverse engineering, robot localization and navigation, object recognition, quality of product measurement, virtual reality, military affairs and so on.General theory and methods of multiple view geometry have been learned firstly, then the paper goes deep into the research of all steps in 3D reconstruction from uncalibrated image sequences, and mainly on two key techniques: feature selection and tracking, implicit surface reconstruction. The paper is divided into the following four parts:1. Feature selection and tracking. The improved feature selection algorithm and optical flow tracking algorithm using iterative Lucas-Kanade method in pyramids are studied, then these two algorithms have been implemented.2. Camera self-calibration. The paper has applied the tracking tool 'boujou' developed by 2d3 company to camera calibration, and implemented self-calibration on image sequences through compiling script file.3. Implicit surface reconstruction. Radial Basis Functions (RBF) implicit surface reconstruction algorithm has been studied to implement surface reconstruction.4. According to 3D reconstruction flow, the main steps above are joined together to carry through simulation experiments of two image sequences, and a feasible scheme for 3D reconstruction from uncalibrated image sequences is implemented, which attempts to solve the problem of reconstruction 3D model from image sequences. The results show that prospective experimental results are obtained for a rich texture 3D scene.
Keywords/Search Tags:3D reconstruction, image matching, feature tracking, camera calibration, RBF
PDF Full Text Request
Related items