GPU-Based Shape from Silhouettes

In this paper, we present a new method for surface-based shape reconstruction from a set of silhouette images. We propose to project the viewing cones from all viewpoints to the 3D space and compute the intersections that represent the vertices of the Visual Hull (VH). We propose a method for fast traversal of the layers of the projected cones and retrieve the viewing edges that lie to the surface of the VH. Taking advantage of the power of Graphics Processing Units (GPU), the proposed method achieves a real-time full reconstruction of VH rather than rendering a novel view of the VH. The experiments on several data sets, including real data, demonstrate the efficiency of the method for real-time visual hull reconstruction.