Researchers develop AI vision tech capable of recognizing irregular behavior to save people

By Park Sae-jin Posted : November 17, 2021, 15:20 Updated : November 17, 2021, 15:20

[Courtesy of ETRI]

SEOUL -- South Korean researchers have developed an artificial intelligence vision technology that uses a closed-circuit television camera system to monitor the movement of people to see if there are any irregular activities such as collapsing. When an abnormal situation is detected, the system will alert its operator. The smart camera system will be installed in the central city of Daejeon for demonstration in June 2022.

AI vision technology enables an AI system to use optical devices including cameras and sensors in a similar way to which a human uses eyes. Based on AI deep learning technology, an AI vision system will actively monitor its environment, recognize objects, and analyze information to check for anomalies.

The Electronics and Telecommunications Research Institute (ETRI) said in a statement on November 17 that its research team has developed "DeepView," an AI vision technology-based system that uses CCTV cameras to detect people in need in real-time. While many AI vision camera systems cannot detect stationary objects such as a person who remained curled up in pain in the street or lying down unconsciously on the ground, DeepView can autonomously find whether a person needs help even if he or she is staying very still.

Some 55,000 image datasets and some 90,000 high-quality information datasets about the movement and position of human joints were used to pump up the AI's cognitive capabilities through deep learning. When the AI system detects abnormal activities, it will quickly contact police and emergency crews.

"We will contribute to the creation of safe and clean urban environments by using ETRI's AI vision technology that can detect abnormal activities with high accuracy even though people's postures are in incorrect shapes," ETRI researcher Min Ok-ki was quoted as saying. DeepView be upgraded by collecting data.

Copyright ⓒ Aju Press All rights reserved.