Apple today has been granted a patent for a method of face detection at different distances in a dynamic picture or live video feed.
The patent, 9,589,177, published on March 7 by the U.S. Patent & Trademark Office, describes a method of “capturing a depth map and an image of scene.” Conventional facial recognition systems scan for faces using multiple “candidate windows” of varying sizes. This not only increases the processing power used, but can also result in an increase in false detections.
To counter that, Apple proposes using an infrared light to project an “optical radiation pattern” onto a given scene. This optical pattern would then be converted into a depth map for a facial recognition system. Put simply, this allows a facial recognition system to detect faces in a live video feed even when the faces are located at different locations and depths within a scene — something that becomes increasingly complex in a live video or a scene that’s dynamic.
To a facial recognition system, human faces become “larger” or “smaller” based on their distance relative to the camera, which can become a problem for the scanning windows. The system described in the patent would use intelligently scaled windows that are tweaked according to their depth coordinates — cutting down on computing overhead and increasing facial recognition accuracy.
Interestingly, the patent also speaks of a “depth mapping and image processing” system, which is likely based on a technology developed by 3D imaging and motion capture startup PrimeSense, which Apple acquired in 2013. While not a biometric recognition or authentication system itself, the method described in the patent could be a small piece of a larger suite of systems and modules for a “revolutionary” front-facing camera on this year’s flagship iPhone. Specifically, Apple’s iPhone X could feature advanced 3D facial and iris recognition systems enabled by PrimeSense’s proprietary technology.