Starting with a set of calibrated photographs taken of a scene, voxel coloring algorithms reconstruct three-dimensional surface
models on a finite spatial domain. In this paper, we present a method that warps the voxel space, so that the domain of the
reconstruction extends to an infinite or semi-infinite volume. Doing so enables the reconstruction of objects far away from
the cameras, as well as reconstruction of a background environment. New views synthesized using the warped voxel space have
improved photo-realism.
Acknowledgments We are indebted to Panoscan for acquisition of the panoramic images of Stanford’s campus, and to Mark Livingston and Irwin
Sobel for calibration of these images. We further thank Mark Livingston for developing a viewer that renders a reconstructed
scene using volumetric warping. We would also like to thank Steve Seitz and Chuck Dyer for numerous discussions regarding
voxel coloring. Finally, we thank Fred Kitson and Ron Schafer for their continued support and encouragement of this research.