Interactive 3D Modeling from Multiple Images using Scene Regularities

Heung-Yeung Shum, Richard Szeliski, Simon Baker, Mei Han, and P. Anandan

Abstract

We present some recent progress in designing and implementing two interactive image-based 3D modeling systems. The first system constructs 3D models from a collection of panoramic image mosaics. A panoramic mosaic consists of a set of images taken around the same viewpoint, and a camera matrix associated with each input image. The user first interactively specifies features such as points, lines, and planes. Our system recovers the camera pose for each mosaic from known line directions and reference points. It then constructs the 3D model using all available geometrical constraints. The second system extracts structure from stereo by representing the scene as a collection of approximately planar layers. The user first interactively segments the images into corresponding planar regions. Our system recovers a composite mosaic for each layer, estimates the plane equation for the layer, and optionally recovers the camera locations as well as out-of-plane displacements. By taking advantage of known scene regularities, our interactive systems avoid diffcult feature correspondence problems that occur in traditional automatic modeling systems. They also shift the interactive high-level structural model specification stage to precede (or intermix with) the 3D geometry recovery. They are thus able to extract accurate wire frame and texture-mapped 3D models from multiple image sequences.

Details

Publication typeInproceedings
Published inEuropean Workshop on 3D Structure from Multiple Images of Large-Scale Environments (SMILE)
Pages236-252
AddressFreiburg
PublisherSpringer Verlag
> Publications > Interactive 3D Modeling from Multiple Images using Scene Regularities