Improved Hardware-Accelerated Visual Hull Rendering

The visual hull is an efficient shape approximation for the purpose of reconstructing and visualizing dynamic objects. Recently, rapid progress in graphics hardware development has made it possible to render visual hulls from a set of silhouette images in real-time. In this paper we present several new algorithms to improve the generality and quality of hardware-accelerated visual hull rendering. First, a multi-pass approach employs texture objects and the stencil buffer to enable the visual hull rendering algorithm to deal with arbitrary numbers of input images. Secondly, flexible programmability of state-of-the-art graphics hardware is exploited to achieve smooth transitions between textures from different reference views projected onto visual hulls. In addition, visibility problems with projective texture mapping are solved by using the shadow mapping technique. We test our rendering algorithms on various off-the-shelf graphics cards and achieve real-time frame rates