Image-based 3D Model Reconstruction For Regular Shape Object

Three-dimensional reconstruction based on 2D images is one of the most activeresearch areas in computer vision and digital photogrammetry, for its promisingutilization in the field of robot vision and virtual reality. This thesis concentrates onthe study of the key problems that are in the process of 3D Model reconstruction forthe regular shape objects, especially for the objects made up by planar surfaces. And afeasible technical method is proposed for the full process of 3D reconstruction. Themain contents of this thesis includes: digital camera calibration algorithm which usesthe focal length stored in the image files, 3D reconstruction of eight point algorithmbased on two views and multiple views, texture extraction and rectification.Camera calibration is a prerequisite for extracting precise three-dimensionalinformation from images. This thesis presents a simple method to calibrate theintrinsic parameters of zoom-lens digital cameras. This method combines the classicalcalibration algorithm using a planar pattern and the focal length stored in the imagefiles. And after a pre-calibration, the mathematic model of the relationship betweenthe zoom lens settings and the intrinsic camera parameters is established. Then we canget the intrinsic parameters correspond with the zoom lens setting of new image filesshoot by this digital camera. Our experiment shows that the proposed method isconvenient to use and can provide accurate intrinsic camera parameters for all the lenssettings continuously.The ordinary eight-point is very vulnerable by the noise in the process of 3Dreconstruction. This thesis investigates the normalized eight-point algorithm andfulfilled an implementation of it for 3D reconstruction by using two calibrated images,and then the method to get a complete 3D model from multiple images is alsoachieved. The iterative algorithm of nonlinear optimization is investigated to getbetter result.In order to get a more photo realistic 3D model, the texture should be extractedfrom the origin photos and be mapped to the 3D model. This thesis presents themethod how to rectify the image plane to get rid of the image distortion byperspective projection.