SPOTR™—Target Recognition

Whether the source is surveillance footage or a digital photo, SPOTR™ uses advanced facial recognition technology to identify people.


Automatic face detection scans the detected person, or the input image, to locate not only the face but the direction the person is looking. With this information, the system can automatically apply an active appearance model, detecting the facial feature points necessary to create 3D models of the target.


Recognition improves when the subject is looking in the same direction as his matching gallery image, but this cannot always be controlled due to surveillance scenarios or uncooperative subjects. Instead of controlling the target, we create a synthetic model of their head and turn it to a more desirable position. 


Not only does the 3D rotation improve automatic recognition, but it helps operators by extrapolating missing information. The ability to rotate the target’s head to different positions will assist any manual recognition tasks.  A witness to a crime, for example, would be able to rotate a target’s head and see them from a frontal pose.


The system improves recognition performance by detecting and controlling the head pose. The head is rotated and compared against our gallery using a spectrum of comparison techniques. This multi-probe, multi-algorithm approach improves performance and allows more accurate recognition.

News

Successful Demonstration of SPOTR™ at Trident Spectre 2012 in Ft Story VA.

SPOTR™ program featured in Wired, Slashdot, and SmartPlanet articles.

Successful Dry Run Demonstration of SPOTR™ Solution for the Office of Naval Research—Princeton, NJ