Research Of3D Face Reconstruction Based On Stereo Vision

In the field of Stomatology auxiliary diagnosis, the face3D visualization and itsresults can clearly record the changes before and after edentulous shaping, this will beconducive to the implementation of a real-time diagnosis, preoperative andpostoperative forecast. The oral field at the present stage, however, the measurementand analysis methods are required as a medium, the plaster model, this method hasmany drawbacks, such as high cost, time-consuming, large systematic errors, dentistsurgent need a fast, high precision face3D visualization which is able to displaythree-dimensional images of the real human face and its texture information, and this isbecoming a computer application research of Stomatology trends. In this paper, theresearch focuses on human face3D visualization, and propose a fast dense disparitymap generation method based on triangulation.In this dissertation, we start with the binocular stereo theory, the entire facereconstruction process is divided into image acquisition, camera calibration, featureextraction, stereo matching and3D reconstruction. The beginning introduce the basicprinciples of stereo vision, and then analyze the imaging model of binocular stereo,Zhang’s camera calibration method and stereo rectification process. Which is confirmedfeasible, efficiently and high precision through experiments, and provide accuratesource data for subsequent matching.This paper focuses on stereo matching of face images, according to the smoothingand low texture features, a fast dense disparity map generation method based ontriangulation is proposed. Our approach firstly match rectified images, build strong setof corresponding points, and then triangulate the points, and to build a linear systemassigns initial disparity, lastly minimize the energy function in the polar directiongradually refinement to yield a dense disparity. Moreover, our method is effective toavoid the disadvantage of easily trapped in local minimum by the global energyfunction optimization, due to the addition of a strong match corresponding point, it aid reduce the match ambiguity. Experiment show that the method is fast and high precision,fully able to meet the needs of practical application.As a consequence through the work, a prototype system of3D face visualizationhas been developed by using VS2010on Windows XP platform and the third-partylibrary such as OpenCV, OpenGL and Vcglib. The system implements the collection,storage, reconstruction and analysis of the face image, moreover, it provide an image ofthe2D display and the3D reconstruction results display. At last, the test results showthat the system in terms of time and accuracy have reached the design requirements, canprovide a simple and convenient visualization platform to the dentist.