Look deeper

Could intelligent cameras and better facial recognition technology hold the key in the fight against terrorism?

For the 2002 dystopian sci-fi movie Minority Report, director Steven Spielberg assembled a crack team of architects, researchers and computer scientists to envisage what the world might look like in 2054. Their predictions proved eerily prescient. Multi-touch interfaces, e-paper and crime prediction software similar to that portrayed in the film has already been developed by technology companies and adopted by a wide range of users including, in the latter instance, police forces in the US.

One technology from the film that hasn’t quite made its way into the mainstream just yet is the ability to carry out retina scanning on individuals at any given time in any given location.

The elite ‘Pre-Crime’ unit deploys hordes of tiny robotic arachnoids to scurry around buildings, temporarily immobilising citizens and scanning their eyeballs to confirm their ID.

Although the technology underpinning this idea isn’t widely available today, law enforcement and anti-terrorism units around the world are already increasingly relying on intelligent cameras and facial recognition software in an attempt to stop would be perpetrators in their tracks – or track them down after the event.

And while video analytics are not being used to thwart crimes before they happen – and will not be used in this way in the future – facial recognition software is becoming more sophisticated. So how effective is the technology underpinning it and what issues need to be overcome before the recognition techniques used in films become a daily reality rather than a science fiction fantasy?

In brief
1. Facial recognition technology is widely used by law enforcement and anti-terror agencies.
2. Powerful new deep learning models and big data have accelerated improvements to facial recognition technology in recent years.
3. The major hurdle developers face is how to combine, in real time, different technologies to improve the accuracy of facial recognition.

Lab to street

The technology underpinning facial recognition has been around for some time, with researchers doing exploratory work in this area as far back as the 1960s, but it’s in the last three years that significant improvements have been made thanks to powerful new deep learning models and the availability of big data, according to Professor Dacheng Tao, director of the Centre for Artificial Intelligence in the University of Technology Sydney.

“This technology is walking out of laboratories and being utilised in many real-world applications in both entertainment and public security,” says Professor Tao.

“To name a few, Facebook and Picasa use face recognition to automatically tag users’ friends in uploaded photos; Sydney airport has adopted advanced customs clearance systems to automatically verify passengers’ identity based on face recognition technology; and police officers in Chicago make use of face recognition to identify a robber’s identity in surveillance videos.”

In the past, similar technology has also been used to thwart voter fraud in the Mexican presidential elections and by the Chinese government during the Beijing Olympic Games to identify activists. But where the technology is expected to have a greater impact in the future is in the area of law enforcement and counter terrorism.

19: The number of people with minor criminal records that were potentially identified by facial recognition software at Super Bowl XXXV in 2001, according to media reports.

Florida police successfully used facial recognition software at Super Bowl XXXV in 2001 to identify potential criminals and terrorists attending the event. Media reports claim that 19 people with minor criminal records were potentially identified by the software. In the post-9/11 world, enforcement agencies and governments have been reluctant to confirm whether and how they are using the technology but anti-terror experts suggest its use is fairly commonplace.

Eric Moncet, head of sector, in charge of citizen security business at Thales, says there are two key ways facial recognition technology is being used by law enforcement and anti-terror agencies: for real time detection and post analysis, following a serious crime or terrorist attack.

In ‘fluid’ situations like the coordinated terror attacks that we have witnessed across Europe in the last few years, the technology clearly offers numerous benefits. However, there are a number of hurdles that need to be overcome before the technology can be widely rolled out. One lingering issue surrounds privacy. In October 2016, a report produced by a think tank at Georgetown University in the US urged for greater controls around the use of images searchable by law enforcement agencies using facial recognition technology. The report said that agencies, including the FBI, had access to a database of more than 117 million American citizens, many of whom were law-abiding and a disproportionate number were black.

Providers of facial recognition solutions have also got to get to grips with a series of technical challenges to make the technology more fool proof, says Professor Tao.

“First, the appearance of the face often changes dramatically due to a number of factors, such as variations in pose and illumination,” he explains. “The appearance variations degrade the performance of existing face recognition algorithms. This challenge is particularly serious for large-scale recognition, where intra-personal and interpersonal appearance changes are subtle. Second, the degradation in image quality in many real-world applications lowers the amount of effective information for recognition; therefore, not all face images can be utilised for reliable recognition.” He adds that facial recognition technology developers are close to overcoming the first problem by designing more powerful deep learning models and collecting more data in terms of different facial appearance variations, which makes it all the more important for the industry to find a way to resolve the second issue.

Florida police successfully used facial recognition software at Super Bowl XXXV in 2001 to identify potential criminals and terrorists attending the event. Media reports claim that 19 people with minor criminal records were potentially identified by the software. In the post-9/11 world, enforcement agencies and governments have been reluctant to confirm whether and how they are using the technology but anti-terror experts suggest its use is fairly commonplace.

Eric Moncet, head of sector, in charge of citizen security business at Thales, says there are two key ways facial recognition technology is being used by law enforcement and anti-terror agencies: for real time detection and post analysis, following a serious crime or terrorist attack.

In ‘fluid’ situations like the coordinated terror attacks that we have witnessed across Europe in the last few years, the technology clearly offers numerous benefits. However, there are a number of hurdles that need to be overcome before the technology can be widely rolled out. One lingering issue surrounds privacy. In October 2016, a report produced by a think tank at Georgetown University in the US urged for greater controls around the use of images searchable by law enforcement agencies using facial recognition technology. The report said that agencies, including the FBI, had access to a database of more than 117 million American citizens, many of whom were law-abiding and a disproportionate number were black.

Providers of facial recognition solutions have also got to get to grips with a series of technical challenges to make the technology more fool proof, says Professor Tao.

“First, the appearance of the face often changes dramatically due to a number of factors, such as variations in pose and illumination,” he explains. “The appearance variations degrade the performance of existing face recognition algorithms. This challenge is particularly serious for large-scale recognition, where intra-personal and interpersonal appearance changes are subtle. Second, the degradation in image quality in many real-world applications lowers the amount of effective information for recognition; therefore, not all face images can be utilised for reliable recognition.” He adds that facial recognition technology developers are close to overcoming the first problem by designing more powerful deep learning models and collecting more data in terms of different facial appearance variations, which makes it all the more important for the industry to find a way to resolve the second issue.