A Single-Shot and Real-Time 3D Imaging Technique for Facial Motion Capture Based on Triple-Frequency Color Fringe Projection
Xiang Zhou,Tao Yang,Zhuangqun Yang,Hong Zhao,Aadrian G. Podoleanu
DOI: https://doi.org/10.15221/12.247
2012-01-01
Abstract:Fast and high-accuracy 3D imaging is playing an important role in scanning human face sunder dynamical change. This paper presents a single-shot technique for real-time surface profiling and texture retrieving of human face. In this technique, three fringe patterns with a carrier frequency ratio of 1:3:9 are encoded in the red (R), green (G) and blue (B) channels of a color fringe pattern and projected onto the object’s surface by a 3-LCD projector. The deformed color fringe pattern is captured by color CCD from another angle. The cross talk among color channels of commercial CCD and projector is modeled as the color coupling equations. Then FFT-Spectrum minimization and bi-dimensional empirical mode decomposition (BEMD) are employed to solve the equations, thereby extracting the low, medium and high fundamental frequency components from R, G and B patterns. 2D-FFT is performed to retrieve wrapping phase distributions of three components. Afterwards, a three-step phase unwrapping strategy is formulated to recursively unwrap phases from low, medium and high carrier patterns. Unwrapped phase of the high-frequency component is selected to recover the object’s height distribution. Apart from the shape, textures of objects are as well retrieved by solving the color coupling equations, facilitating 3D real construction combined with textures. The algorithm is accelerated by GPU parallel computation, fulfilling so far 1 frame/s 3D real reconstruction.