Capture of hair geometry using white structured light

Abstract

Structured lighting (SL) scanning technology has been successfully applied to reconstruct personalized avatars, providing high-quality geometry for the face and body. However, previous white SL methods have typically been unable to capture and reconstruct hair geometry, preventing a complete, realistic avatar surface from being reconstructed. In this paper, we propose a novel hair capture system, producing a full-head manifold with complete hair, based on four white SL scanners made by ourselves. The key technical contribution is a robust strip-edge-based coding algorithm, using a projection pattern of 18 stripes; it allows geometric acquisition to an accuracy of 1 mm. Experiments are given to show that our system produces high-quality hair geometry, comparable to that captured by more expensive state-of-the-art multiview stereo systems. Our system is however easier to configure, and is more suited to real-world setups.