文摘
Capturing a face in a multi-user collaborative environment has been a problem of concern for several years. The problem becomes more severe in mobile applications where the capturing system may occlude the user's field of view. In this thesis, a system is proposed that captures two side views of the face simultaneously and generates a frontal view. Applications of facial capture system include tele-conferencing, wearable computing, collaborative work and other mobile applications. The system is designed to produce in real-time a stable, quality frontal view of an HMD user whose face is captured with little obstruction of the field of view. The frontal views are generated by warping and blending of the side views after a calibration step. In tests, the generated views compared well with real video based on both normalized cross correlation and the Euclidean distance between some of the prominent feature points. Preliminary qualitative assessment of these generated views also concludes that the generated video is adequate to support the intended applications. A 3D face model that can support the generation of arbitrary views was also constructed.