Image-based 3D Reconstruction From Multiviews And Texture Rectified

It is a hot point in Computer Vision and Digital Photogrammetry that using digital images to extract texture and reconstruct the corresponding 3D model, and has a vast application future in some fields such as robot vision, 3D measure and so on. Under the direction of computer vision principle, this paper focuses on the critical problems of extracting texture and 3D model reconstruction particular those which are composed by plains, and explore a feasibility method to reconstruct 3D models and rectify texture based on images which have some overlapping feature taken by digital cameras.Perspective distortion of the texture which is extracted from images must be removed before mapping it to the model. But the common software that can deal images only use mouse to change shape of the image, but can't modify the shape exactly accord your need. This paper provides another way to rectify texture which is simple and fast based on single view reconstruction. The image which is generated can be used for texture mapping in 3D modeling.A 3D model can be reconstructed from two views if the corresponding camera is calibrated. Based on the principle of reconstruction from two views, the process of getting homography matrix is improved, and better effect can be obtained after mapped the rectified texture which origins from the images to the part models reconstructed. Following, the principle to generating the whole model by every double views is deduced based on the principle of reconstruction from two views and is realized. The result is pretty good.Reconstruction based on three views are studied and trifocal tensor are computed and the corresponding foundation matrix and camera external matrix,which can be used to resumed 3D model with camera internal matrix.