Researchers developing "soft biometric" video analysis system

Researchers in Australia are developing a way to identify individuals using “soft” biometrics like their estimated weight, hair color, and skin tone in video footage.

The researchers hope to create a Google-style search, where police officers can actually search for an individual in hundreds of hours of video footage just by typing in a basic description.

“If you want to search for something on Google, you type in keywords and you look for it. It is a similar concept that we want to do for people on video,” explained Clinton Fookes, an associate professor at Queensland University of Technology’s (QUT) School of Engineering Systems.

“If there is a mugging somewhere and the description of the person is six-foot tall, wearing blue jeans, a black shirt and a red baseball cap — we can take that description and convert it into an algorithm to actually data mine video networks to search for a person of that description,” he said.

Current video analytic software is capable of detecting break-ins, unattended baggage, and interpreting actions, but still unable to identify individuals.

Fookes hopes that the new soft biometric video analytic system will be able to assist law enforcement agencies find suspects more quickly.

“The London bombings, for example, I think they had about 40 or 50 federal agents looking at video for a couple of weeks full-time to manually locate what happened, who were the people that did it, where did they come from, where did they go and tracing them through the video.”

The technology is still under development phase and researchers are working to improve the prototype’s accuracy.

“Anywhere from the order of 70 to 95 per cent accuracy is possible, depending on the challenges of the environment,” Fookes said. “In the next 12 to 24 months, we are looking at doing some testing in live environments with operational systems.”