Soon,a supercomputer that recognises face,gait,human activity

A suspect is fleeing a crime scene through a crowded marketplace.

Written by Johnson T A | Bangalore | Published: June 27, 2013 1:16 am

A suspect is fleeing a crime scene through a crowded marketplace. Closed circuit cameras discreetly scanning the crowds register his speed of movement and trigger alarms. Meanwhile,even as other cameras keep tracking the suspect,a computer system,which the cameras are linked to,pick up a match in terms of the suspect’s face and gait from a database for crime suspects. It sends out a real time alert to the law enforcement agencies.

What is described above is a futuristic scenario,but with the application of video and closed circuit camera technology on the threshold of a quantum leap of pervasiveness,researchers are chipping away at machine vision technologies that could become the norm for security in modern cities.

At the only video analytic laboratory in India — Supercomputer Education and Research Centre in the Indian Institute of Science — scientists have,as a part of a DRDO-funded security and surveillance project,come up with technologies that enable computers to recognise faces,walking styles and even human activity from video data with up to 90 per cent accuracy rates. Using computer vision modules,video compression cues and computer technology,scientists at the lab have tested algorithms that can help search reams and reams of video footage within seconds,creating technologies that are akin to Google searches for the video world.

With cities like Bangalore and Hyderabad set to install CCTV systems as security measures under the Central schemes like the Mega City project,evolving video surveillance technologies are likely to play prominent roles in urban security infrastructure.

A team of scientists,led by Assistant Professor R Venkatesh Babu,at the IISc’s video analytic lab have in the last three years come up with over half a dozen scientific papers on feature matching,movement tracking and action recognition from video sequences with funding from the DRDO under its ‘Compressed Domain Human Activity Analysis’ project.

“Visual surveillance and security applications were never more important than now,more so,due to the overwhelming threat of terrorism. The major objective is to remove humans from the loop and see if computers can see and understand like humans. We are really far away from this objective,but we are moving towards it. Video technology has become an important part of modern human lives. Mobile phone cameras have great resolution and are getting cheaper. All cities are going for camera surveillance. So cameras are there and huge data is coming in and all we need is the tools to do things on that,” Babu says.

One of the fundamental things that the lab has done for video surveillance is develop technologies that dramatically reduce search time for specified information from the terra bytes of data that normally get stored during video surveillance. “If an event happens people go and search through huge piles of video data,up to one terra byte. This can take days. We work in the compressed domain and use motion information to drastically reduce search speed. But we need to know what to look for — like a car moving at a particular speed — to pull out matching info from the video meta data,” Babu says.

According to him,intelligent video surveillance is an “essential security provision in high-vigil areas such as airports,railway stations and sensitive establishments”.

The emergence of the Kinect camera that provides 3D images and depth information has helped some of the video analytics research work,but with these cameras not suitable for outdoor scenarios,many new security applications like action recognition are restricted to indoor scenarios,Babu says. He also warns that technologies are far from being fully usable in real life situations.

For all the latest News Archive News, download Indian Express App

    Live Cricket Scores & Results