Now boarding: Better ways to count airport crowds

Modern airports seem to be awash with advanced technology and data — security cameras, cell phone tracking systems, biometric scanners — yet managers still struggle with a deceptively simple question: How many people are actually here right now?

The answer has implications for how smoothly the airport operates and, ultimately, the passenger experience.

In real-time, managers must decide when to deploy staff, what wait times to communicate to travelers, and whether the airport is approaching capacity limits. They also need to run long-term planning scenarios for adding flights or making operational changes.

To address these challenges, Professor Adam Mersereau and his colleagues set out to build mathematical models that – paired with people-counting sensors – estimate crowding in the security area without visual headcounts.

“We saw an opportunity to solve two problems at once,” he says. “Passengers would get accurate wait time estimates so they could time their arrivals better, and the airport would get better data for staffing and long-term planning.”

Mersereau worked with Serhan Ziya at UNC-Chapel Hill, Morgan Wood, at North Carolina Wesleyan University, and Duke University’s Fernando Bernstein and N. Bora Keskin.

Using Raleigh-Durham International Airport as their test case, Mersereau and his fellow researchers, found that while sensors in TSA areas make mistakes that accumulate over time, smart algorithms can correct the errors. They also found that airports can predict when security lines will be busiest by analyzing typical passenger arrival patterns for flights.

Their research has applications beyond airports, potentially helping managers at theaters, hospitals and concert arenas better anticipate and manage crowds.

The difficulties of counting the crowd

Mersereau’s expertise centers on what keeps brick-and-mortar retailers operating smoothly. He’s especially interested in the murky realities of inventory and in-store operations — the stuff that isn’t captured by sales data. Unlike e-commerce sites, physical stores are more like a black box: Managers might know what’s selling, but not who’s coming in, what they’re doing or whether the store is adequately staffed.

Airports pose a similar kind of problem, he says. Managers don’t necessarily know how many people are in a space at a point in time.

“It’s not just passengers waiting in lines, but also bags, planes and crews,” says Mersereau. “And with physical queues, it’s mostly guesswork. No one’s standing there with a clipboard tracking line length in real time.”

To come up with an accurate count, Mersereau and the team installed infrared beam sensors at the entrance and exit of the TSA area at the airport. Each time a passenger breaks the beam, the system logs either an entry or an exit. At any given moment, it tracks the total number of people who have entered since the start of the day and how many have left. The difference between those two figures can, in theory, be used to estimate how many people are currently there.

The trouble is the system’s counts of people coming and going are inherently “noisy.” A couple walking arm-in-arm might register as one person. A big rolling suitcase might count as two. A person inadvertently leaning on the sensor can throw it off completely.

“These little errors add up, and they do so in ways that eventually make the estimates pretty useless,” says Mersereau.

The remedy lies in an algorithm that involves strategic resetting of population estimates. When the system can confidently detect when exits slow or stop through patterns in the departure stream, resetting the count to zero prevents errors from building over time.

The algorithm is most effective at smaller airports like Raleigh-Durham, where traffic naturally rises and falls — say, an early morning rush at 6 a.m. and a lull two hours later at 8 a.m., according to Mersereau. In high-traffic airports where crowds never let up, additional data, such as occasional manual counts by airport personnel, can improve performance beyond what the algorithm alone achieves, he adds.

What’s your show-up profile?

In a separate study, the researchers took on the challenge of forecasting when people arrive at the airport. The goal was to help managers strengthen both short- and long-term planning and scheduling.

There are many traditional forecasting methods, from classical statistical approaches to newer machine learning and AI-based models, but many operate without much transparency. “You feed them data and they spit out a forecast, but they don’t take into account what’s unique about airports.”

Airports, says Mersereau, are structurally different from most environments. After all, people don’t just show up randomly; they come because they have a flight to catch. “In other words: This is a different kind of forecasting problem because there’s structure to how people arrive. Making sense of that structure can make us better forecasters.”

A key concept in forecasting airport arrivals is what’s known as the “show-up profile,” which maps out how early different passengers typically arrive for their flights. Conservative travelers usually show up two hours or more before their domestic flight; others tend to push it and arrive just an hour before.

The challenge is that airports have no way of knowing exactly which passengers are arriving for which flights. TSA scans boarding passes, but it doesn’t log arrival times due to privacy reasons. Without that information, understanding these show-up profiles is difficult since it’s impossible to know who’s who.

To solve this, the team devised an algorithm to estimate show-up profiles for each flight, predicting when passengers arrive based on limited data. Their estimates reveal a number of patterns.

For early morning flights, passengers tend to arrive closer to their departure times; in the afternoon, they’re more likely to build in a buffer. Business travelers often cut it closer than leisure travelers, and those connecting through major hubs typically arrive later earlier, too.

This kind of forecasting is useful for short-term workforce scheduling throughout the airport. “If we can predict when people will arrive at the TSA, the parking garage and even the shops and restaurants inside the terminal, managers can better plan when and where workers are needed,” he says.

This forecasting approach also helps leaders with long-term planning. Since passenger arrivals are tied to flight schedules, airports can use it for “what if” analysis when considering adding or rescheduling flights.

Mersereau’s research has applications beyond TSA checkpoints and airports. It could help managers at venues with scheduled events, such as stadiums, hospitals and museums, better predict and manage crowds.

“Nobody likes dealing with long lines — not travelers, not TSA agents, not the barista behind the terminal coffee counter,” he said. “Everyone has something to gain from making the whole system more efficient.”