designing out crime
Face-to-Face Value
Facial Recognition Technology Is Ready, But When Will It Become a Permanent Feature?
First impressions count, as the saying goes, as we all too often make instant judgements about people from their appearances. Their non-verbal communication—features, postures, and the clothes that they wear—speaks volumes in communicating impressions of their personalities. Nineteenth century psychiatrist and criminologist Cesare Lombroso went one step further and built his professional reputation upon a theory—now discredited—that a proclivity towards crime was written into someone’s physical DNA and that we could determine intent by simply looking at their features.
Now in the twenty-first century it can be said that we are revisiting a strand of the theory through the use of technology—facial recognition (FR), providing LP professionals with the ability to use cameras and software to accurately identify and distinguish between legitimate customers and troublesome individuals with whom LP has had previous encounters.
Although not Lombrosian in pre-determining criminal intent, it singles out known criminals through their facial characteristics captured on a store’s surveillance cameras. This seemingly futuristic capability is possible through the use of cutting-edge biometric matching software whose algorithms can render a suspect in vivid 3D and then look for matches in a store’s database of known criminals. This technology is the next big step in the marriage of CCTV technology and business analytics. Why then has FR experienced such a difficult journey to being accepted at face value?
As a technology that has rapidly gained traction in a post 9/11 era, FR has come a long way in the last few years. Previously the exclusive tool of Police, anti-terrorist units, and Government intelligence departments for the purposes of preventing the acquisition and use of fake identities, FR is now experiencing significant uptake in the commercial world as a natural progression. And not surprisingly, retailers have been watching this evolution with interest.
Rather like RFID technology, it has its uses beyond security, for example in marketing and upselling to loyal customers, but this retail nirvana could only be realised when the accuracy and the price were right.
In the LP world, certainty of identity is paramount so as to make sure that politely asking a person to “leave the store” does not blow up in their faces through some mistaken identity and brand damaging social media campaign.
Indeed, getting it right could have huge benefits for retailers and change the relationship with law enforcement in an environment of public sector cuts. In the UK, Police forces have endured a 20 per cent reduction in service across the board with more cuts imminent, and similar stories are prominent across Europe. They simply cannot respond to every shoplifter call out.
In a perfect world where FR technology is cost-effectively employed across an estate, a person identified as someone who is “not welcome” would simply be asked to leave as they attempt to enter the store. This efficiency will save LP staff time, which would otherwise be spent trying to locate, observe, intervene, apprehend, detain, and turn a suspect over to Police. If one store has this technology, then it is simply displacing the issue to another store, but if every retailer in a shopping centre employed it, FR could have huge time and cost savings and avoid the need for lengthy prosecutions.
How Does It Work?
Someone caught stealing is stopped by the store detective or security guard who is trained in making an approach, and they advise the suspect that as a condition of not pressing charges they must enrol in the store’s LP system.
This enrollment takes the form of recording biographic and incident data as well as a facial picture. The suspect would then be advised that they are persona non grata and now prohibited from entering any other stores. With that understanding and agreement, they are free to leave without further action. By consenting to do this, the enrollee’s face is logged into the database (called a gallery), which can be shared across the rest of the store estate. In this way if they attempt to enter another branch, that store’s FR system picks them up, and they are once again asked to leave. Additionally, by using smartphone technology a notice can be sent to all LP associates to heighten awareness of a possible threat.
In this respect, it is a totally “opt-in” experience rather than a “capture-all” technology, which is important to remember when we later refer to the privacy arguments circulating around the FR debate.
Cost savings have even been quantified in terms of time. Ideally the simple threat of immediate identification by the FR system will act as a sufficient deterrent. Simply approaching a suspect as they enter the store takes about three minutes. Compare this to watching a suspect go around a store and waiting for them to steal, which takes on average twenty minutes. If they are detained, that involves an average of forty-five minutes of their remaining in the store. The total time saving here per case is a minimum of one hour of a store detective or security guard’s time. It is therefore easy to make a financial case for technology making LP more cost and time efficient.
Quality, Cost, and Accuracy
Up until now, one of the inhibitors to widespread adoption of FR in retail environments has been the quality of the video images. Previously, the grainy images of customers captured on predominantly analogue systems have been so poor that even when LP staff were watching an event as it happened they were hard pressed to make a positive identification. By contrast, today’s high-definition IP cameras offer crystal clear images well suited for FR. Factor in the latest software that can convert a 2D video frame into a 3D model and accurately map the contours and textures of a face, and you have a powerful system that can perform matching in the most difficult of situations.
There seems to be two separate issues at play hindering the final hurdle to wider adoption. Firstly, although many accept that the technology is proven, it has until recently been seen as expensive to install.
The second more thorny issue—although it is had not halted the FR adoption rates—is that of privacy.
The perceived expense of FR is not the matching software per se, but the installation of dedicated FR cameras and integration with other store technology like video management systems, local area, and virtual private networks, according to Jack Ives of US software developer CyberExtruder who, over the last 16 years, has built the software that makes robust face matching from video possible.
“System cost was a blocker. But the precipitous decline in price of high-definition cameras and the new low-cost HDCVI cameras have significantly lowered the bar and accelerated adoption,” said Ives.
“Cameras that were once costing thousands of euros can now be bought for hundreds. Labour to put the whole system together is still the larger cost centre, but as integrators become more accustomed to the new paradigm, that cost is falling as well. Previously a bird’s eye view of the store entrances was the desired setup. Now, when the goal is face matching, the cameras need to be located lower in order to capture people’s features.
“One of the other early issues was that people are milling about, and the camera was only able to capture rough angles including side views, which was not conducive to matching. That is why we have developed a process of making the image a 3D model, which now provides the infinite ability to re-pose a facial picture and then carry out accurate matching.”
Access control and even replacing computer passwords with FR technology will form part of the future to reduce the chances of cyber attacks or hackers accessing personal information.
Ives said that although many companies are still using analogue networks, the HDCVI camera technology produces a solution that marries a low-cost HD camera with legacy analogue cabling and is FR capable.
“We have been trialling this with a couple of businesses, and the feedback is interesting because the benefit has proved greater than we all expected,” he added.
Privacy Is a Matter of Geography, Not Technology
The other potential obstacle to FR technology has been privacy, although as Ives stated, “Privacy is part of the conversion but has not stopped the adoption.”
Putting it into context, it is a geographical issue rather than a technological challenge. In Europe, where data protection rules are enshrined in member state law, there are no real obstacles because customers implicitly consent when they are made aware that they are subject to surveillance for security reasons through easy-to-see signage. So there is implicit and assumed consent as soon as they enter a store.
Also, for LP purposes, because the store thief voluntarily enters an agreement to be photographed, they have also consented. And only images of suspects are circulated within a closed loop of retailers, so there is a clear understanding of the rules of who has access to the images and why.
In the US however, there is no such assumption, and privacy groups have been trying to work out a code of practice or standards by which to protect consumers from any “abuses” of FR scans.
However, in June 2015, after sixteen months of protracted talks, privacy advocates—of which there were nine groups involved—walked away from the negotiating table saying that they had “hit a dead end” in getting agreement from the retailers to protect shoppers from facial scans.
This may delay acceptance of FR in the US because the issues the campaigners are arguing about are less tied up with store thieves and more to do with the focus of retailers and holding data on legitimate customers. However, it will not halt the roll-out of LP applications.
According to Jürgen Pampus of Cognitec Systems, which has been working in the FR space since 2002, the issues can easily be resolved. He said, “Depending upon the country, there are quite strict privacy regulations that users of FR products have to follow. But these restrictions can easily be addressed by applying the main privacy rules whether it be information affecting clients or the restricted use and storage of that information.”
He also said that the use of FR for marketing purposes is relatively new, which is probably why it is not so clean cut in terms of applying the rules governing privacy. This has muddied what is a clear picture as the two protagonists—the retailers and the privacy groups—seem to be arguing at cross purposes.
Alvaro Bedoya, a law professor at Georgetown University and leading privacy advocate, said that consumers should have to give permission by default before they could be scanned. But retailers argue that people previously convicted of shoplifting would not voluntarily give permission.
Privacy spokeswoman Juliana Gruenwald said, “At a base minimum, people should be able to walk down a public street without fear that companies they’ve never heard of are tracking their every movement—and identifying them by name—using facial recognition technology. Unfortunately, we have been unable to obtain agreement even with that basic, specific premise.”
The biggest concern among privacy groups is use of the technology by retailers to target and profile people. So long as a company has an existing photo of “persons of interest,” from shoplifters to “your best customers,” staff can be sent an email or text alerting them of that person’s arrival.
Facebook has been using facial recognition for a few years now to identify and categorise pictures. It’s easy to see this data being shared among retailers and social media sites to target ads for users. Facebook already knows what products are looked at online by scanning cookies from retailer websites and uses the results to bombard users with ads for the products they just looked at or already purchased.
They argue there is also the danger of the data being compromised because although customers can change passwords and credit card numbers, they physically cannot change fingerprints or the precise dimensions of their faces.
Interestingly, Facebook has decided not to offer its photo-sharing app Moments in Europe because of regulator concerns over its FR technology.
It is a challenge for LP professionals that have been held back from keeping out the bad guys by bigger privacy arguments that do not apply to their sphere of operation. But the facts remain that we live in a more surveillance-led world because of the ongoing threat of terrorism, and privacy campaigners are more likely to become nervous at more draconian use of so-called “big brother” technology.
For example, the Electronic Freedom Foundation believes the FBI already has 14 million face images on its database and plans to increase this to more than 50 million. In the UK, the Police have about 18 million mugshots on file.
Just two states in the US—Illinois and Texas—have adopted Europe’s approach, so it could be a slower adoption of FR in the US than in continental Europe. However, even with the data protection rules, the campaigners argue that the holders of data do not always play by the rules of engagement and hide behind the blanket of national security when challenged.
Once seen as science fiction, FR technology is rather like Minority Report pre-crime deterrence. Once the privacy issues have been resolved, FR has the potential to change the rules of retail by lowering shrink and optimising the time of the LP team as well as the Police and criminal justice system. It has the potential to lower prosecutions while at the same time deterring the crime in the first place. FR is now part of the topography of crime mapping—the reading of individual features to provide a route map to their intent as they enter stores. For those with dishonest intent, retailers hope that FR will make that journey and store visit a short one.
