The proposed research aims to extract 3D facial models for recognition applications using passive imaging sensors and computer vision techniques. We propose exploiting image pairs acquired simultaneously using a stereo camera rig, as well as short image sequences from single cameras. We currently provide commercial stereo products that perform well under cooperative, controlled enrollment conditions, and seek to extend the technique to dynamic illumination, longer ranges, and near-real-time performance for probe imaging. Using single-camera sequences, we have demonstrated initial 3D acquisition capability, and we now seek to achieve robustness and performance envelope improvements sufficient to support operation in realistic deployment scenarios. In Phase I of the proposed effort we seek to 1) implement initial robustness and processing time enhancements to support characterization work, 2) establish and archive a database of subjects acquired using both methods, 3) characterize baseline performance envelopes and accuracy compared to laser scanner data, 4) validate use of acquired 3D shapes in a 3D recognizer, and 5) document potential designs for Phase II development. The ultimate technical goal is to acquire facial shapes rapidly and robustly under realistic operational conditions and process them in 3D facial recognition systems that discriminate reliably despite varying pose and illumination. The proposed effort will develop enabling technology that should ultimately yield sensors capable of recognizing people in a manner similar to the way their friends do - extremely accurately, non-intrusively, and immediately. Such sensors could be used to control access to facilities, continuously monitor operator workstations, confirm identity at border crossings, match air travelers against wanted lists, and otherwise support a wide range of military operations, force protection, and homeland security applications. Revenue potential is in the hundreds of millions of dollars.
Keywords: Facial Recognition,Human Identification,3d Imaging,Shape,Biometrics,Access Control
