S'pore: NTU develops AI surveillance software
Developed by Associate Professor Maylor Leung of Nanyang Technological University's School of Computer Engineering, and his PhD student Li Liyuan, the cameras combine artificial intelligence software, which allows them to differentiate between "normal" behavior--people walking and talking--and when things go wrong.
After two and a half years of research, they have trained the system's "neural network" to tell when someone falls, fights, or breaks furniture.
The system has a 96 per cent record for detecting abnormal behavior, based on simulations using actors. It has given no false starts. It chalked up the impressive results by considering 80 shape and motion features from live camera footage, among them shape of a moving object, its speed, position and direction--all described in a mathematical formula, since computers do not see like people.
Better still, the neural network will get smarter with practice.
The surveillance system "would clearly be for security, to watch out for fights and robberies in airports, shops and so on," Prof Leung told The Straits Times. "It can also be used to keep an eye on elderly people at home."
The researchers are interested in working with industry to develop the system into a commercial product, which could cut down on manpower costs. "One security guard could be watching over 100 monitors," noted Li.