FAR – Facial Attribute Recognition Jim Austin Advanced Computer Architectures Group, University of York Cybula Ltd.
FAR project • Concerted action with AICP (costs and people shared) • Started 1 August 2003 for 2 years. • FAR Collaborators – Cybula Ltd. – QinetiQ – University of York
People • University of York – Dr. Nick Pears, Academic manager – Prof. Jim Austin – Dr. Mike Freeman, Technical and Amadeus Link • Cybula – Dr. Nick Walton, Technical Lead. – Dr. Sujeewa Alwis, Commercial manager • QinetiQ – Dr. Andy Lewin, Technical Lead.
Motivation • Many applications where recognising a face (or hand etc.) would be valuable. • Present systems are expensive, large and power hungry • A low cost camera, aimed primarily at face recognition would be highly beneficial. • Cybula/UofY has a system that this can be based on
Aims • To build a low cost 3D camera to primarily support facial biometrics. – Selection and evaluation of camera technology – Development of embeddable camera software • Camera calibration • 3D image generation
Technology Cybula Camera CM-2 Stereo Lenses Projector Texture lens
Technology • QinetiQ camera – Special spot projector – Very low power (USB power) – Potential low cost
Reconstruction Stereo camera 1 Objects Stereo camera 2
Typical Images
Image correction • Radial Distortion correction After Before
3D reconstruction • Next stage – Finding correspondences in the two images – Helped by the spot projector – Must be as fast as possible – Must fit on DSP/FPGA as a core
Conclusions • Camera technology identified • First problem solved – camera calibration • Next stage is 3D reconstruction • Port to hardware
Recommend
More recommend