We propose to use 3D stereo-based sensors to provide passive, fast, compact, 3D capture of the human face. During Phase 1 we investigate the feasibility and performance of this method with an experimental capture rig for cooperative subjects. The device consists of one to three pre-calibrated stereo camera heads installed in a compact rig, about the size of a briefcase. Depth sensing will be performed by the Tyzx DeepSea stereo-vision ASIC. Depth sensing results will be merged into a3D model for validation against reference objects. These results will be used to specify a deployable stand-alone, single view sensor unit with small footprint, low power, capable of providing real time results. We will also investigate the utility of using this system to capture 3D image sequences which can be of value to recognizers and include this data with a 3D image database acquired with the capture rig. In Phase 2, the standalone unit will be constructed and evaluated under a wider set of operating conditions. Face recognition and identification capabilities are severely limited by variations in pose. Emerging research indicates that using special techniques combined with 3D data of subjects may significantly enhance face recognition performance. Contemporary 3D facial image acquisition, however, is cumbersome and time consuming, requiring multiple sensors, complex apparatus, and cooperative subjects. Frequently, these techniques make use of active sensors (laser scanning or structured light) which is objectionable to subjects. These factors render the collection and use of such imagery unsuitable for most military and civilian real-world applications which require a low cost, compact, self-contained sensor unit with minimal or no setup process. The ability to use a single sensor unit to take a 3D "snap-shot" of the face represents a breakthrough with respect to ease of application within existing security infrastructures and the opportunity to significantly enhance face recognition performance. With inexpensive acquisition of 3D data available, recognizer performance will be enhanced and 3D enrollment and recognition systems will become used in port of entry applications, access control, customer relationship management and potentially for credit and bank cards.
Keywords: Human Identification, Face Recognition, 3d Face Imaging, Head Tracking, Stereo Vision, Surveillance, Cmos Image Sensors, 3d Sequence
