Electronic surveillance in the city is likely to undergo a marked transition in the coming years, from being reliant on humans to being driven by machines, courtesy research that is under way at Indian Institute of Technology, Hyderabad (IIT-H).
In a paper published in the journal Pattern Recognition Letters , IIT-H researchers Debaditya Roy and his guide Krishna Mohan Chalavadi have described a model for a system which they claim can detect snatch thefts, like chain or purse snatching, in video footage.
“The dependence on video surveillance is on the rise. In the present day, archived footage has to be scanned by a cop in a control room to look for something specific. An automated system would save both time and human effort. That apart, such a system can be scaled up to detect events real-time,” said Debaditya Roy, the study’s corresponding author.
Roy, who is pursuing Ph.D at the institute, explains crime in a surveillance video constitute a very small part, with a bulk of it being non-significant recording. This makes it tedious for a human to search. By generating what is called an action vector, Roy’s system attempts to see patterns in the movements of people captured in a video, which in turn helps pinpoint a theft.
“Our model has been trained with dataset from Hyderabad police. It has also been trained with data comprising actions or patterns that resemble snatch theft,” he added.
For two years now, the researchers at IIT-H have been working to develop systems that can identify thefts, accidents, and also identify wanted persons in video surveillance footage. Research on such futuristic systems began after the Hyderabad police signed an agreement with IIT-H in 2016.
Roy says his model has shown efficiency greater than the existing state-of-art models. The next step is to scale it up to see if it can detect thefts in real-time, live footage.
If the systems being modelled at IIT, Hyderabad, show real-time efficiency, they could soon find their way into police control rooms and greatly reduce the response time after a crime has taken place.
