Three Dimensional Surface Reconstruction Based On Image Sequence

With the development of the computer techniques, it is not satisfied any more by people to see simple two dimensional scenes in the screen. Hence, three dimensional reconstruction techniques have attracted more and more attention and have been widely used in the fields of computer aided design, computer vision and virtual reality. In this thesis, the octree based 3D reconstruction algorithm is firstly described and then algorithms for surface points extraction and for texture attachment are proposed. The main work and contributions of this paper are summarized as follows:Firstly, combined with the conventional octree based 3D reconstruction algorithm, a 3D surface reconstruction algorithm is proposed. In this method, the vertices of the cubes composing the object are classified to six cases and different weights are assigned to vertices according to their different states of being a vertex or in an edge or in a face of a cube. Each vertex is scored according to the numbers of vertices, edges and faces it occupies. According to the scores corresponding to the vertices, they are classified to the internal points and the surface points.Secondly, a contour extraction algorithm is proposed for the planar contour reconstruction. In this method, parallel planes are used to intercept the cube for obtaining the interception points, according to the characteristics of which the edge points are extracted. If the cube that the edge points belong to is connected the points can be judged to be connected too and they can be classified to one contour. Otherwise, those points will belong to different contours. The number of contours of each slice and also the correspondence relationship between points can be obtained after classifying the points according to different contours they belong to. The planar contour can be obtained after linking those points sequentially. As compared with the method that direct triangulation is performed with respect to the surface points, the problem of some triangular elements being in the inner part of the object can be well resolved by the extraction of the planar contour, which is very helpful for texture attachment.Finally, the shape and appearance of the object are restored with the texture mapping method. The texture image is selected according to the angle between the view point (also called the normal vector of the image) and the normal vector of the triangular element. That is the image with the smallest angle will be selected as the texture image corresponding to the triangular element considered.