Three tech companies took part in a facial recognition trial at Philadelphia International Airport, aimed at replacing manual checking at security gates. Carroll McCormick discusses the results with the facility’s Allen Mehta
Bn etween January and mid-March 2020, Philadelphia International Airport (IATA: PHL) evaluated biometric facial recognition systems supplied by the Metropolitan Washington Airports Authority (MWAA), NEC and SITA at three international gates. The photos taken of passengers before boarding their flights were compared with images stored in government databases, thereby obviating the need for manualchecking of passengers’ passports.
In 1996, the United States signed into law the Illegal Immigration Reform and Immigrant Responsibility Act. It was, however, only fully implemented when the US President Donald Trump signed an executive order that the programme be fully rolled out by October 2021.
In short, the Act requires that, in the case of airports, photos be taken of arriving and departing international passengers and compared with photos held by the Department of Homeland Security, such as for passports or other travel documents. The primary objective is to eventually replace the manual checking of passports, but PHL’s evaluation programme also sought to demonstrate that the systems could serve as a noncontact replacement for the manual checking of boarding passes as well.
Allen Mehta, acting chief information officer at PHL, noted: “The way the US Customs and Border Protection [CBP] has defined it, airports need to do this for the entry process, [but] the exit process needs to be working first. It’s tied to the same system, but CBP wants it to be working for the exits for international flights.”
The division of labour being proposed is that airports and their airline partnerships are responsible for the facial scanning equipment for departing passengers, while the CBP is responsible for the equipment for arriving passengers, according to Mr Mehta. However, he added: “CBP is part of the partnership, as we will not proceed without their input, such as on meeting regulations and future additional needs.”
PHL tested the three vendors’ systems at three of its 22 official international gates. The original plan was to rotate the systems around the three gates, allowing each to be tested as equally as possible. “What we were looking at was where we could get the best bang for the number of flights: Cancun, Canada and the Middle East. Variety was one of the major criteria,” Mr Mehta said.
The reason why we chose three different vendors was so we could test different technologies, but at the same time we wanted to make sure that our security needs would be addressed and meet industry standards
Allen Mehta acting chief information officer, PHL
However, once COVID-19 spread to the US, the project had to end a week early. During the seven weeks of testing, about 15,000 passengers on 107 flights were processed. “We had sufficient data to evaluate the equipment and get reasonable evaluations,” Mr Mehta said.
The scope of the project was to do a comparative analysis of the three vendors’ systems using a two-step biometric exit process. First, take photos of passengers’ faces for comparison with stored photos. Second, manually check passengers’ boarding passes. PHL had planned to run one-step processing tests, too but, as Mr Mehta explained: “Integrating of the boarding pass and facial ID, planned for the final weeks, had to be dropped because of COVID.”
The airport completed its report early this summer and has shared the results with CBP. The data gathered has provided feedback for the system manufacturers, while simultaneously helping the airport to write its Requests for Proposal to the potential suppliers of the permanent biometric facial recognition systems installed in the furture.
PHL intended to evaluate the systems on 13 criteria, but because the pandemic cut the trial short, the airport was unable to gather data on four criteria: ease of integration, functionality (does the equipment operate as expected), ease of operation and ease of use of software. It rated the other nine criteria – product availability/production delivery, cost, aesthetics (appeal), ease of deployment, ease of configuration, match performance (speed and accuracy percentage), downtime/problem resolution/ error reporting, quality of support (if applicable) and operational reporting features – on a scale of one to five, with one being the lowest and five the highest. One vendor scored a total of 34 points; the other two each scored 39 points.
Mr Mehta explained: “The reason why we chose three different vendors was so we could test different technologies, but at the same time, we wanted to make sure our security needs would be addressed and meet industry standards.”
Digital signage was set up to display the instructions on how to use the systems and, because they were programmable, they could be tailored to the flights.
Mr Mehta continued: “We have a lot of passengers whose primary language is not English – for example, on flights to Qatar. We were able to put instructions in Hindi and other languages. That was one of the great successes we had. As people were sitting in the hold rooms, they could read the signs and ask questions before boarding began, so as not to delay the boarding of the flights.”
The biometric process was straightforward and quick. Passengers simply walked up to a spot marked on the floor and the cameras took their photos. In less than three seconds, on average, the device would clear passengers.
The resulting match/no match varied by supplier. For example, passengers and gate agents might see the word ‘YES’, allowing passengers to proceed to the agents’ desks and show their boarding passes, or a ‘NO’ in the cases of a failure to match the photo with a stored photograph. Or, they might see a green tick to proceed and a red ‘X’ indicating a failure to obtain a match.
Resolving match failures was easy, Mr Mehta said: “The current process is that, if the biometrics rejects the face recognition, the gate agent will ask the passenger to present their passport and boarding pass as they would have done before. Gate agents can ask a CBP agent to help if they suspect anything. Customs may come up and do spot checks and assist, as necessary.
“All we did was refer back to the original boarding procedures if the faces didn’t match or if there was a technical issue. To my knowledge, during the test period only one individual required extra checking, but the person was cleared and released. A few passengers – mainly US citizens – declined the face recognition process. We simply checked their passports and they were cleared to the planes.”
The project team observed passengers for a couple of days to determine the best place for the footprint decals that showed passengers where to stop and then approach the cameras. “We were able to learn [both for PHL and its airline partners]. This is one of the reasons we did this project, so we could see what would happen. Each airport is different,” explained Mr Mehta.
As long as we had power, we could have gone wireless if we wanted to.
What were some of the things PHL learned? Ease of deployment, for instance, varied between the manufacturers. As Mr Mehta observed: “NEC and SITA had a standard unit: plug in the power and ethernet cable and it was good to go. MWAA [the VeriScan system] had an extending arm that had to be attached to the agent counter, and it took a little while to decide where to place it. We had to look at where to place the cables so no one would trip over them, and if we moved the units, not to stretch the cables.
“The equipment was light enough to be easily put into place. [Some of them had] wheels. Even though all the vendors had Wi-Fi connect capability, we required that they work with ethernet cables. As long as we had power, we could have gone wireless if we wanted to.
“The only real challenge was with ease of configuration – how the software connected to the internet. For example, with open ports you must make sure that [the necessary] port is turned on and communicating with the internet or CBP. The main challenge was in making sure that the networks were connected so that the information could go to where it needed to go to.
“It really wasn’t a big challenge – more a lack of communication with the vendor teams, such as a vendor technician asking to open a port and us wanting to protect the environment, [and ensuring that no one was] able to hack in. We needed to document the procedures.”
The systems achieved an overall match rate of 97-98%, and where the usual average speed for manually processing a passenger is eight to nine seconds, Mr Mehta confirmd: “We got responses in less than three seconds.”
He explained how they arrived at the 4/5 and 5/5 scores for accuracy: “We had a lot of discussion on that. Take a single flight with 100 passengers. Give 5/5 if you can get 98-99% accuracy without a whole lot of false positives or negatives. If that could not be met, give a 4/5.”
One issue relating to downtime/problem resolution/error reporting was due to dropped internet connections, such as losing the link to the CBP site. An issue that occurred with powering up the devices was traced to a manufacturer doing updates at night and the new version not downloading because of a lost connection. “The maximum downtime for any vendor was maybe 30 minutes. We will want to build in redundancies – for example, a cellular connection as back-up. We will have these requests in the RFP.”
One operational reporting feature that required some extra work was figuring out how to place the MWAA movable arm so passengers would not bang their heads. The lack of a movable camera with the SITA product was considered a drawback. “There is only one physical configuration. Since it is a fixed height, a tall person would have to crouch down to be photographed. The SITA system didn’t have the capacity on the back end so the agent could raise the camera for the capture,” Mr Mehta said.
We did see some funny cases where people saw their faces and wanted to fix their hair.
“We had one person who was close to seven feet tall. Crouching was not a good option for him. There are advantages to having an articulating arm, like zooming the camera up. It’s easier for one person, but it takes time because it makes it harder for the next person in line. There are some cameras that automatically move to people’s faces. The technology is there and the request will come up in future requirements.”
He notes that passengers can see their faces in the screen, with a predictable result: “We did see some funny cases where people saw their faces and wanted to fix their hair.” Tactful gate agents kept the passengers moving.
What about masks?
Another likely future requirement is that facial recognition equipment be able to identify faces even when passengers are wearing face masks, as has become increasingly commonplace during the COVID-19 pandemic.
The airport team observed that proper lighting was important. “Good natural lighting yields better matches then when there are shadows. It can easily be solved with better background lighting. Lighting is something that airports need to think about. Vendors need to think about that; for example, installing small LED lights,” Mr Mehta explained.
A pro for the NEC system was the easy-to-navigate user interface and a two-camera option that facilitated photographing passengers in wheelchairs: “This was important to us because this is not an IT person operating this equipment. We want the instructions to be simple enough for the agent to use.”
An pro for the MWAA was the Commercial Off The Shelf (COTs) iPad it used as the camera/computer: “In general, we like COTS equipment because you can replace parts quickly, but there are also restrictions. For example, with Apple IOS you cannot do remote connections into iPads to do support. If there is a technical difficulty and a time crunch, the only [solution] is to swap out the equipment. There are advantages and disadvantages of COTS and proprietary equipment. You need to keep some spare equipment.”
MWAA was the only supplier that did not allow remote access for vendor support: “We asked the vendor the question and they said to just to get another iPad. It takes a few minutes to load the software and you are ready to go.”
The installation process also formed part of the evaluation, with one team considered to have fallen short.
Although some people suspect that the US government’s intentions with facial recognition may well go beyond their advertised aim of simply expediting airline passenger processing, it is a fact that in this new world of global pandemics, contactless passenger processing can help with limiting the spread of viruses.
“In today’s world we are trying to go touchless,” Mr Mehta concluded. “This is going to be a big help. With this technology we could have done one-step processing – clearing you to go on the plane. This is an advantage we can apply to this realm that we weren’t thinking of in January when we set up this project.”
In general, we like COTS equipment because you can replace parts quickly, but there are also restrictions.