Kaist creates tool that can automatically detect facesResearchers from the Korea Advanced Institute of Science and Technology have developed a facial recognition system based on a computer chip that can process faces with the lowest amount of electricity possible, allowing it to effectively run continuously.
Prof. Yoo Hoi-jun and his team at the university’s electric engineering department said Tuesday that they have developed a product called K-Eye that comes in two variations: a device that can be worn around the neck and one that can be attached to a phone’s charging port for use with a smartphone.
Equipped with a camera, K-Eye devices are able to automatically detect a human face, recognize it on a database and recall the subject’s information including his or her name.
The core feature that makes K-Eye distinct is its ability to constantly detect faces using low electricity consumption of less than one milliwatt. This means instead of the user having to manually turn the device on, K-Eye can continuously recognize the presence of any human face in front of the wearer and turn on automatically.
K-EyeQ, the variation that can be attached to a phone, can immediately show a subject’s personal information on the smartphone as soon as a human face appears in front of the device.
The team said the two technologies behind the power-saving are its “Always-On” image sensor and CNNP semiconductor chip.
The Always-On image sensor is capable of reducing standby power by automatically turning the device on when it detects a human face.
CNNP is a semiconductor that can facilitate machine learning, in which a computer is able to process information without explicitly being programmed to do so. The chip received recognition at the International Solid-State Circuits Conference, a semiconductor conference, in February for consuming the least amount of electricity in the world.
There is great promise in hardware like CNNP because most artificial intelligence technologies introduced by global tech companies, such as Google’s AlphaGo, run on software instead of semiconductors. The disadvantages of software are that it is slow and consumes a lot of power, points that can be overcome by CNNP.
CNNP has a success rate of 97 percent in facial recognition and runs on electricity of 0.6 milliwatts, about a 5,000th of the energy consumed by AlphaGo’s graphics processing unit.
BY SONG KYOUNG-SON [email@example.com]