Tomorrow’s robots and security systems could be able to face-recognize three times as many people using a fraction of the processing power, according to NEC. The company’s new large-scale image processing tech can track and identify faces and clothing – as well as potentially suspicious behavior – for three times the number of people in-frame simultaneously, compared to existing systems, a potential boon not only to security installations in airports and shopping centers, but for your friendly home robot.
Facial recognition in robotics so far has generally been a compromise between power, accuracy, and speed. The open-source Qbo robot, for instance, uses a combination of local and cloud-based processing, able to learn and the recognize familiar faces and objects, as well as crowd-source identification of objects other Qbo ‘bots have been trained to identify.
However, early home robots – such as Sony’s Aibo – were altogether more humble in their capabilities; the robot dog, for instance, was able to spot and track a colorful ball, but its face recognition was rudimentary. Similarly, identification systems for phones and laptops, where systems can be unlocked after recognizing the user, have struggled to balance the responsiveness those users demand with the accuracy required for true biometric security.
NEC’s system could streamline that process, by reducing server load for tracking smaller numbers of people at any one time. The primary implementation is likely to be in security systems, however, where the more efficient monitoring will allow fewer staff to keep on top of greater numbers of potentially dangerous people.