Using facial recognition to process surveillance footage isn’t a new concept — Scotland Yard began using the technique in the wake of last year’s London riots — but a new system developed by Hitachi Kokusai Electric could make the process quicker and more flexible than ever before. Shown off at this year’s Security Show expo in Japan, the system processes footage from both still images and live footage, recognizing faces in real-time, and can search through up to 36 million different faces for a comparison in one second. Recognized faces are displayed as thumbnails and grouped with matching footage, allowing the operator to look at all of the actions of a given person immediately.
To achieve the speed, the system implements the facial recognition portion of the process while the footage is being recorded from the cameras, rather than waiting to process the imagery after the fact. That said, there are some trade-offs being made: the system only works with faces turned within 30 degrees of the camera either horizontally or vertically, and faces must be at least 40 pixels in height and width for the system to work properly. Hitachi Kokusai Electric plans to begin selling the system in its next fiscal year.