Protect the safety of citizens with Visual AI

Data set consisting of human domains, detailed joint positions and postures for a visual AI deep learning study. Credit: Electronic and Telecommunications Research Institute (ETRI)

Korea’s research team from the Electronics and Telecommunications Research Institute (ETRI) announced that it has applied visual AI technology “DeepView” to the metropolitan city of Daejeon to prevent safety accidents in the city and respond quickly.

DeepView is AI technology that recognizes human behavior. It detects people lying on the road in real time using surveillance cameras. As it can be applied to the prevention of safety accidents caused by alcohol, fainting, etc. and the implementation of rapid emergency rescue measures. It is expected to become the basic technology to make a city safe.

Most of the behavioral recognition technologies out there have a two-step structure of first detecting a person and then recognizing their position. Therefore, they had problems not correctly detecting a person with atypical postures, such as crouching or slouching, compared to a person standing.

DeepView recognizes a person’s behavior, recorded by a surveillance camera, by analyzing the detailed information of 18 joint points and 6 postures. By developing the optimal deep learning model that understands and interprets the correlated data simultaneously, the research team dramatically improved the recognition rate of atypical postures and shortened their detection time.

ETRI explained that the core technology accurately recognizes human behavior by simultaneously using human domains, detailed joint position, and posture data.

In other words, it automatically detects whether or not a person has actually fallen in a specific place by developing a model that simultaneously takes into account many deciding factors rather than recognizing the behaviors step by step.

The research team could increase accuracy by using in-house constructed image data that contains over 55,000 human cases and a high-quality data set comprising over 90,000 cases of human domains, detailed joint positions and postures as well as the deep learning study.

When a person falls, an immediate real-time response is crucial for safety. DeepView detects a fallen person in real time in conjunction with the surveillance camera control system and notifies them to the control center.

ETRI has been applying this technology to real examples in Daejeon since June. When DeepView expands nationwide, as noted above, it will make a significant contribution to citizen safety by solving control blind spots through extended surveillance.

Ok Gee Min, ETRI Deputy Vice President, said: “We will continue to help build a safe and comfortable city with ETRI’s visual AI technology that quickly recognizes abnormal behaviors with great accuracy, even if the person’s posture is not normal.

AI-based visual technology for use in CCTV cameras

More information:
The research team announced the technology in the AVSS 2021 program on November 19: the basic dataset and models to detect human postural states in a robust way against irregular postures (Document 71)

Provided by the National Science and Technology Research Council

Quote: Protecting Citizen Security with Visual AI (2021, December 17) retrieved December 18, 2021 from

This document is subject to copyright. Other than fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for information only.

Comments are closed.