University of Kentucky
As the technological capabilities of terrorists become more sophisticated along with their resolve, identifying such individuals within a free and open society is an issue of international prominence. The identification of select individuals from the general population requires a system that, while efficient, is also innocuous such that a person is scanned with no more user cooperation than by having that individual enter a public building, board a plane, or drive through a police check point. This transparency requirement greatly limits the application of biometrics that rely heavily on user cooperation to be practical. The most promising technologies that address these security concerns are based on face recognition where a user’s identity is determined by a comparison of that person’s photograph, taken at the time of the ID, with photographs stored in a database. But making a positive match with a high degree of accuracy between photographs, taken under varying lighting conditions, head poses, and facial disguises is proving to be a very difficult problem.
Here at the University of Kentucky, we are developing a structured-light video camera that extracts 3-D depth information from a scene in real time for cockpit interfacing, and we believe that the integration of the depth information that this system provides with existing 2-D image based face recognition technologies could dramatically improve the accuracy of existing face recognition systems. Some of the obvious gains obtained with depth include a third dimension of measurements for feature extraction and the ability to compensate for differences in pose/lighting. Real-time depth information could also dramatically improve scene segmentation algorithms by creating a broad array of new criteria by which to isolate objects such that faces can be identified by shape as well as color. The subject of our research is to build a transparent surveillance system based on the integration of RGB and real-time structured light video to address both obstacles of scene segmentation and face recognition without cooperation.
For more information, see: http://www.engr.uky.edu/%7Edllau/Research/surveillance.html