Research On Texture Mapping From Photograph

Virtual reality is an important subject in the field of computer graphics and computer vision, and it has a wide range of applications in cultural relics preservation, special effects on movies, advertising, e-commerce and other fields. Usually,to the 3D reconstruction of real scenes, it will enhance realism through the method that adding texture for the 3D surface model. The process usually takes two steps constructing three-dimensional geometrical model and adding texture for the geometric model. With the developing of photoelectric range finder technology, it has become reality that reconstructing the shape of an object by the 3D scanner. Using digital photos rebuilding objects texture is the key points of the texture mapping research.The texture mapping based on digital photos is that using digital cameras gain actual object surface texture image, and then establish the mapping relationship between 3D model and texture image through perspective projection matrix, mapping the true texture to the surface of 3D model to enhance realism. This paper has a research on the key technologies of the texture mapping based on digital photos, including the following:(1) Introduced the development situation of texture mapping technology at home and abroad, Recountted the theory knowledge related on texture mapping technology, including computer graphics, computer vision, camera calibration theory, image registration method and some other relevant knowledge:(2) In order to solve the calibration of camera extrinsic parameters problem in texture mapping,this paper presents a geometric calibration method based on four point pairs.The first one selects three points in the world coordinate system and calculate their coordinates in the camera coordinate system(world points, light heart, and image point are collinear). Transforms the coordinates to a reference coordinate system and the euclidean distance is invariant. According to the homothetic triangle theory, calculates the rotation matrix and translation vectoer respectively. Finally, gets the complete extrinsic parameters..Applied in the texture mapping, this method can establish a mapping between 3D model and digital photo:(3) The key points of the registration method based on 3D-2D feature points is finding the matching points between 3D model and 2D image. This paper adopts man-machine interactive way of selecting corresponding feature point, and puts forward a fast 3D pick-up method based on the perspective projection. It can pick up the three-dimensional space coordinates;(4) Aiming at the perspective change of the texture images, after research the SIFT features registration algorithm, using Mahalanobis distance eliminate the mismatch points of the SIFT algorithm to improve the accuracy of image registration.