Face Recognition Technology has become a leading tool in alerting for known rogues in major public security situations, NEC Europe head of Global Face Recognition Solutions Chris de Silva informed a gathering of some 35 of its Australian customers.
Speaking at a gathering to promote NEC’s Technology in Torquay, Victoria on 24th October, de Silva was unashamedly proud of NEC’s continued strong test results against his global competitors. Most recently, its facial recognition technology led a group of 15 other systems in National Institute of Standards and Technology testing (NISTIR 8173).
It gained the top ranking in NIST’s most recent report on Face in Video Evaluation (FIVE) in March. These sought to test how well the various solutions could capture faces in real time of walking subjects. In one test based on scanning individuals as they pass through an airport gate without acknowledging the camera, NIST’s technology bested its rivals with an accuracy rate of 99.2 percent. This compared very favourably to the 95 per cent rate achieved for others. Its system attained an error rate only half as high as the nearest competitor.
Despite this victory, the ability to apply this technology in large real time events, such as a major sports or community gathering was something only possible in James Bond movies, Da Silva said.
“There’s two different scenarios,” he said. “There’s real time and there’s forensic face search.”
Real-time accuracy tends to decline as the population being surveyed increases.
"Technically it’s fine,” he said. “But the issue is that you are going to get false alarms. If the person you are looking at is moving their head – the score is coming down.”
If you have a lot of false alarms, the system will become unusable and its credibility as a deterrent is undermined.
For later forensic non-real-time use, it can be useful and proved an indispensable investigative aid. However the trade-off of false warnings makes sense. However it’s accurate use in a go or no-go decisions in large uncontrolled crowds is not yet available.
“You are going to find more people that look similar as you expand the size. It’s just the nature of human beings,” he told ADM.
De Silva also criticised proposals for upgrading to 3-D captures as a likely solution.
A recent review of Face Technology Trends (2016-2020) concluded that 2D facial recognition, the dominant method used by NEC, had disadvantages due to errors from poor illumination, posture variations, and facial expression variations.
It argued that 3D-facial recognition was more accurate, as it provided enhanced details of an individual's face with sensors that project structured light.
De Silva said this approach played down the cost of setting it up.
“Getting it perfect for a 3D-stereo image all around your head is going to be a challenge,” he said.
He said a 3D-camera was better, in theory, if you ignore costs and need for new infrastructure. Its improvement would be marginal at best and there was an issue of what other use a move to 3-D would assist.
At the same time NEC’s technology worked well with most CCTV feeds which were used for other purposes as well.
“Practically speaking, 3D-face recognition has very limited use cases.” he said.
Disclosure: John Hilvert attended the event as guest of NEC, who paid for his transport and accommodation expenses.