VOL. 67 December 2021
ETRI Protects the Safety of Citizens with Visual AI
- Applied AI technology detects falling people in real-time from surveillance cameras
- Expected to prevent safety accidents in the city and rapidly respond to them.
The Electronics and Telecommunications Research Institute (ETRI) research team developed a technology for detecting humans lying on the road in real-time. Therefore, it is expected that a safer society can be realized by preventing and promptly responding to accidents in the city.
The Electronics and Telecommunications Research Institute (ETRI) announced that it had applied the Visual AI ‘DeepView’ technology to Daejeon Metropolitan City in earnest to prevent safety accidents in the city and respond to them quickly.
DeepView is an AI technology that recognizes human behaviors. It detects people lying on the road through surveillance cameras in real-time. The DeepView technology can be used to prevent safety accidents caused by excessive drinking and fainting. It can be applied to prompt emergency rescue measures and is expected to become the core technology for making a safer city.
Most of the conventional technologies recognize human behavior in two steps: 1) detect a person and 2) estimate the postural state of the person based on the detected region. However, they usually fail on detecting a person with illregular posture such as being crouched or lying down, compared to a standing person with an upright posture1).
DeepView accurately recognizes human behavior by analyzing detailed information on 18 joint points and six postures. The research team developed an optimal deep learning model that dramatically increases the recognition performance for irregular postures and shortens computation time.
1) standing, walking, running, sitting, crouching, lying
ETRI explained that its core technology is to accurately recognize human behavior through multi-task learning, which simultaneously utilizes the human postural states and detailed joint position.
The research team could increase the accuracy by building a high-quality dataset that contains more than 55,000 images and over 90,000 human instances with detailed joint positions, accurate regions, and postures.
When a person falls down, immediate real-time response is crucial. DeepView detects a fallen person in real-time in conjunction with the surveillance camera control system and notifies the event to the control center.
ETRI has been working with Daejeon Metropolitan City to apply this technology since June. When DeepView expands nationwide, it will significantly contribute to the safety of citizens by resolving control blind spots2) through wide-area surveillance.
Ok Gee Min, the Assistant Vice President of the ETRI said, “We will continue making a contribution to building a safe and comfortable city with ETRI’s visual AI technology that quickly recognizes abnormal behavior with high accuracy even if the person’s posture is not normal.”
This technology has been introduced as the AI technology to tackle public social problems in the ITU <AI for Good 2021> and published the formal thesis in the International Conference on Advanced Video and Signal-based Surveillance (AVSS) conference paper. Also, the DeepView project was selected as the Top 100 Excellent National R&D Achievements with its outstanding research this year following 2019.
2) A problem of not being able to monitor at the right places as control manpower is insufficient compared to the number of CCTV videos
