Facial recognition technology applied in trying environments

Facial recognition technology that could be used in low-light environments has taken a step forward thanks to researchers from Defence Science and Technology (DST).

The technology would also be able to identify people from long distances, up to 250 metres.

The research sets the stage for the wider use of facial recognition technology in security products, which despite the wide availability of CCT footage have been hampered by the inaccuracy of facial recognition technology outside of ideal environments.

The two scientists, Sau Yee Yiu and Dmitri Kamenetsky, used an in-house facial recognition algorithm for the research.

The algorithm was updated with insights gleaned from research into how heat propagates in the atmosphere.

“This turns out to be similar to the way noise from atmospheric turbulence distorts images over long distances. The atmosphere moves and shifts around and your image gets sheared and blurry. Applying my heat dispersal model gets rid of that turbulence and brings it back closer to a focused, sharp image,” said Yiu.

To account for the distortion of the image caused by low light conditions, the algorithm incorporates filter passes to remove graininess.

The parameters by which the image can be altered are controlled by sliders which update the image. This functionality allows for the image to be individually altered to account for the particular conditions which it was taken in.

According to Kamenetsky, this research has wide applicability to security and monitoring systems and will be broadly available.

“We’re very happy with the results, which will be of benefit to stand-off surveillance systems,” said Kamenetsky. “We’ve released a description of the algorithm, allowing other researchers to implement it and make further improvements.”

Yiu and Kamenetsky work within DST’s biometrics research team. The researchers at DST aim to track human presence via images or other data such as fingerprints or iris scans against databases of characteristics that are held on file.