SMARTER CCTV RECOGNIZE HUMAN ACTIONS

[Anchor Lead]

CCTVs are becoming smarter and smarter. Not only do they notify of possible dangers by detecting movements or sounds, but recently, CCTV technology that allows the detection of multi-complex actions even in dark or foggy conditions when visibility is poor has been developed by a group of university students.

[Pkg]

A man suddenly collapses in an elevator. The emergency is first detected by a security camera. It sends an emergency alert message to the control room right away. When a group of men are spotted in a fight in the middle of the street, a security camera detects it immediately and recognizes it as a fight. Two post-graduate students from the University of Science and Technology succeeded in the development of a human action recognition CCTV using artificial intelligence deep learning technology. The camera can recognize more than seven kinds of abnormal situations, such as roaming, intrusion, falling down, fighting, abandonment, arson, and marketing.

[Soundbite] Jeon Ho-beom(UST student) : "We used the zero-shot learning method, which enables the camera to infer abnormal situations and recombine information on its own."

The camera can accurately identify human actions even in dark or foggy conditions when visibility is poor. It can even detect dangers such as arson.

[Soundbite] Kim Hyung-min(UST student) : "The camera learns explanations and images of actions such as igniting a lighter or pouring gasoline."

This CCTV has been certified by the Korea Internet & Security Agency, and recorded the human action recognition rate of 94.66 percent in a performance test conducted by Nanyang Technological University of Singapore.

[Soundbite] Prof. Kim Do-hyung(University of Science and Technology) : "It’s the only domestic technology to have acquired certifications for detecting more than seven specified situations."

The researchers say the camera can also detect crowds in certain spaces, and plan to push for its commercialization soon.