Image-based model reconstruction using textured planes

In this paper we propose a representation of image-based models by textured planes augmented to surface geometry reconstructed from a sequence of depth images. Each textured plane is a textured and partially transparent plane into which the texture from the input images is mapped using projective texturing. That is, a set of textured planes, or billboards, that look like the input when seen from the input viewpoints. The process comprises two fundamental steps, repeated until the input data has been exhausted. These steps in our method, are spatial sampling of the input scene and search for fitting planes. In the first step, the geometry of the objects in the input scene is sampled using measures of optic flow. Then, we find a set of planes that fit the approximate sampled geometry obtained in the first step. Under this hybrid reconstruction, texture mapping can be employed to deterministically identify coincident rays from different view points and thus achieve high quality rendering, with taking full advantages of the acceleration utility of graphics.