現在位置: ホーム / 研究紹介 / 実環境のモデリングに関する研究 / Real-time Space Carving using Graphics Hardware

Real-time Space Carving using Graphics Hardware

Christian Nitschke, 画像の認識・理解シンポジウム(MIRU)講演論文集, pp. 928-933, Jul. 2006.

論文概要

Reconstruction of real-world objects from a set of multiple images is a topic in Computer Vision and 3D Computer Graphics with many interesting applications. Several attempts have been made to real-time reconstruction by employing PC cluster systems or a single workstation with computer graphics hardware. While the former provide enough power to obtain good quality in realtime, they are expensive and less flexible. The latter ones achieve good performance compared to single CPU algorithms, but are not applicable in a real-time context. This work shows our efforts in developing an GPU hardware accelerated algorithm, providing an explicit dynamic volume representation at real-time. High quality is achieved through combining silhouette and photometric techniques for determining the colored surface of a 3D object. We explain our approach using programmable vertex and fragment kernels with current graphics hardware and compare it to a highly optimized CPU version. We show, that the new approach can outperform the CPU version several times and give an outlook for promising future enhancements. (pdf file)