Has facial recognition finally come of age? T

remendous challenges face our police and security services when tracking

suspected terrorists, keeping an eye on people living in our country and spotting threats coming across our borders - very often through legitimate routes like air- travel, but using fake documentation.

Humans are naturally very good at spotting people they know, even from small glimpses, but are poor at recognising those unfamiliar to them, so blanketing our transit hubs and public spaces with teams looking for suspects is a fruitless and unaffordable exercise.

Through recent significant advances in facial recognition and other identification technologies, automatic alerting is finally becoming a very practical tool.

Most facial recognition solutions are “one-to- one”, where systems are used to confirm an identity in relation to a presented document – such as a passport or a work security badge. So, the test is fairly simple – is the person standing in front of the camera the same one recorded in the passport? For “one-to-one” facial recognition, there is no need to search against a database of millions of people that have valid passports – just a comparison between two photographs.

This challenge is made as simple as possible by utilising a high-quality passport photograph, good lighting, perfect camera height, and no difficult items covering the face, such as hats, for the comparison picture. Yet, it often still fails. But, if you watch enough Hollywood movies, we are led to believe that governments can find wanted people instantly using low quality street cameras, dispersed across a city.

The leading-edge capability to find a “person of interest” in a crowd, or standing in an immigration line, is a significant leap from the technology used in passport e-gates – it presents ground-breaking capabilities, even if not quite at the level of Hollywood’s imaginations.

We refer to this new type of check as “one- to-many, non-compliant”, meaning that we are searching for the person amongst a larger database of suspects, and they are not presenting themselves in perfect conditions to the camera – in fact, they may not know there is a camera there at all.

A camera watching a potential target area for a terrorist attack is unlikely to have perfect conditions – the resolution of the face will be lower even though we may be using a high


Facial Recognition Technology doesn’t have to be 100% accurate to be useful – just hundreds of times better than the human brain

definition camera because the subject will be at distance, and most probably moving. The face may be in partial profile and without a neutral expression. The subject may be wearing a hat, the lighting may be poor, or their face may be obscured by hair. I am often asked, “How accurate is the potential of facial recognition in these circumstances?”, with an expectation that it will be close to 100%.

However, I like to turn the question round and say how much better does technology need to be than a human brain? Is hundreds or thousands of times better, even if it does not boast 100% accuracy, a useful tool? We are close to the point that facial recognition technology on one camera can out-perform a hundred people watching that feed, and that gives us interesting options.

So what has driven the development of technology used to more effectively find people in a crowd?

Last year there was a lot of noise regarding machine learning being used to assist Google’s AlphaGo program to beat Lee Sedol, the top ranked go-player. New facial recognition technology is using the same machine learning capabilities to improve its detection accuracy, and critically, its speed of detection, to improve performance. It is specifically focused on identifying people in a crowd, using video as a key differentiator, whilst many traditional systems treat video as a collection of still images, and extract the most “passport-like” image from the video to compare against. By using multiple images when tracking a subject, new technology increases its detection accuracy, because each frame gives out new information.

Humans do still have a role – the level of involvement depends upon the criticality of selecting a target from video. If seeking out a white-collar criminal who has missed a court hearing, for example, I would set the system with a high threshold as too many sightings for a low tier case would result in police officers’ time being wasted. But, with strong intelligence that a suspect has travelled to a

www. c i t y s e cu r i t yma g a z i ne . com

major city with a bomb, I’d want to bring all sightings that are fairly close to that person’s likeness to the attention of the police. The ability to dynamically set the “importance” of detection, the relationship between false alerts and false rejections, is critical.

Of course, there are understandable objections regarding privacy and appropriate levels of monitoring. But, it is important to recognise that these systems are not deployed to track “unknowns” – they are simply being used to compare people against suspected persons in a set of existing watch-lists, and sightings of normal public can be discarded if no match.

Lists of people who should be excluded from our borders already exist within our agencies, and most people would likely be much more comfortable if everyone entering our borders received an automatic check performed against those lists.

Cutting-edge technology allows that check to be performed whilst people are waiting in line, prior to reaching an immigration officer, ensuring that wait times and processing burdens are not impacted.

Finally, it is not just governments that can take advantage of this technology. Security managers in our cities need to be vigilant against threats, and many are provided with lists of persons to watch out for.

This same technology can be applied within buildings on CCTV cameras and smartphones to look for excluded persons and raise immediate warnings to security teams.

We are even seeing applications of facial recognition for fast track, touchless entry at building security gates. It looks like facial recognition is finally coming of age.

Mark Patrick Chief Technology Officer Digital Barriers

> 9

Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32