Modeling 3D scenes from video

Abstract

We present a technique to obtain texture mapped models of real scenes with a high degree of automation using only a video camera and an overhead projector. The user makes two passes with a hand-held video camera. For the first pass the scene is under natural illumination, and a structure-frommotion technique recovers coarse scene geometry and textures. For the second pass a grid of lines is projected onto the scene which allows us to acquire dense geometric information. The information from both passes is automatically combined and a final model consisting of the dense geometry of the scene and a properly registered texture is created.