Passive spatiotemporal geometry reconstruction of human faces at high fidelity.

Abstract

Creating photo-real digital human faces remains one of the biggest challenges in computer graphics. A core component of the challenge comes from the high complexity of the human face, and the difficult task of building realistic faces is aggravated by human perception. Based on the work of his doctorial thesis, the author and his colleagues developed a fully passive and markerless system that reconstructs facial geometry at the skin pore level and at a full frame rate, achieving the highest spatiotemporal resolution to date. Passive capture setups are simpler than active ones and only require cameras and static illumination. The author's research has already been implemented in production systems and used by the commercial videogame and entertainment industries.