The world's first Crowd Behavior Analysis Technology that detects congestion status and anomalies from crowd footage!
Recently, more and more security cameras are being installed at stations and on the street. However, with more cameras to monitor, surveillance staff apparently face an increasingly heavy burden. With this in mind, NEC is working on the development of Crowd Behavior Analysis Technology. This technology can analyze crowd footage from security cameras, and send notification only when it detects that something out of the ordinary may be taking place. From what I hear, this technology is a world first! I went straight to visit the leader of development, Mr. Miyano, to find out more!
Interviewee: Mr. Hiroyoshi Miyano
Mr. Miyano is an expert in image recognition. He is currently the leader of development for the Crowd Behavior Analysis Technology, which determines congestion status and detects anomalies based on crowd footage from security cameras, etc., without identifying individuals.
- MitaHello, Mr. Miyano. Getting straight to the point, what exactly is Crowd Behavior Analysis Technology?
- MiyanoTo sum it up, it is technology that determines congestion status and detects anomalies from crowd footage taken on security cameras, etc., but this is probably a little hard to envisage. To make sure everyone gets a good idea of it, can I start by discussing how we came to develop this technology?
- MitaThat would be great.
- MiyanoRight now, a growing number of security cameras are installed at stations and on the street. With more security cameras, those responsible for monitoring them have a higher workload, so we wondered if we could create a system that analyzed security camera footage, and only notified staff when an anomaly occurred.
- MitaI've already heard a bit about that.
- MiyanoWhen developing a system like this, analyzing crowd footage that shows large numbers of people is a challenge. Conventional video analysis involves following each individual, and detecting anomalies from their behavior, but when the location is crowded analysis becomes more difficult. It is tricky to follow each individual, and when people overlap it becomes impossible to differentiate between them. Then there is also the problem of privacy.
- MitaAnalyzing footage that shows lots of people must be hard.
- MiyanoThat's why we thought we'd see if it was possible to analyze crowds in clusters, rather than following and analyzing individuals. In this case, analysis is possible even if a location is congested, and targeting clusters means individuals are not examined, allowing you to take privacy into consideration. That's how we came to develop our Crowd Behavior Analysis Technology.
- MitaI see. How do you analyze crowds as clusters?