INSUBCONTINENT EXCLUSIVE:
Main image: Amazon's Rekognition face ID system
a police procedural, but thanks to improvements in facial recognition tech, real law enforcement officers have begun experimenting with it
In the UK, both London's Metropolitan Police and South Wales Police have used facial recognition systems to pick would-be troublemakers out
of a crowd in real-time at large events, and at the 2017 Champions League soccer final in the Welsh capital Cardiff, police made an arrest
after the system threw up a match with a database of wanted criminals.Critics, however, claim the system is inaccurate, and South Wales
Police statistics showed that 92% of nearly 2,500 matches at the Champions League final were 'false positives'
But others say that number is meaningless on its own and should only be considered in the context of how the system works
systems, Heseltine tells TheIndianSubcontinent, work though deep learning, a type of machine learning that uses virtual neural networks to
essentially reconstruct a simulation of a human brain
You can teach the system to recognize faces by showing it hundreds of thousands of pairs of faces and telling it whether they match or
systems are used by airports, construction companies, and in other venues where security is a key consideration
It all depends on what you want the system to do in practice
whether two images of a face are of the same person; it simply assigns a likeness score based on the similarity in their characteristics
For example, for a camera capturing real-time video, the system could look at a face in a frame of the footage and compare it to all the
faces on a known police database
faces are matched.A police van equipped with a facial recognition camera
considerably, Heseltine explains, depending on the importance of finding a match, the consequences for a false positive, and the man power
For example, if the police were looking to catch a petty criminal, they might only look at the very highest likeness scores
But if they were trying to catch a serial killer they might set the threshold lower, and follow up anybody that the system identifies as a
potential match, given the heightened importance of catching that person.The 'likeness threshold'The best systems are now very accurate,
Heseltine says, and will nearly always give high likeness scores to matching faces if the image quality is good enough
So, why would police systems have high false positives There are several factors at play
threshold' that the police chose to record a potential match
and have a fantastic report if they turned [the threshold] up a notch
ensure real cases are caught
Airports will set the threshold based on the number of passengers they can follow up
but it can struggle when the image quality is poor
The force has since installed special cameras on vans that work better.It can also be difficult for a system to identify people if their
head is at an angle, or if part of their face is obscured by shadow
are working to improve detection in poor quality images, or where part of the face is obscured
In addition to the CSI-style zoom-and-enhance technique, other innovations include systems that can predict what the left side of the face
related to the way police and other organizations use photos and other data linked to the systems will remain, and this week workers at
Amazon demanded that the company stop selling its facial recognition software, called Rekognition, to law enforcement
evaluation of the year-long trial of the use of the technology by South Wales Police is still ongoing, and the results could prove pivotal
to the future use of facial recognition in the UK.South Wales Police and the National Police Chiefs' Council declined to be interviewed for
this article.TheIndianSubcontinent's Next Up series is brought to you in association with HonornXZAjskNuQB4RD34QnddDY.jpg#
