The technologies behind the Internet of Things are offering new ways for managers to keep track of how many customers are using a service — for example, the real-time number of people in a brick-and-mortar space like a grocery store or a bank.

However, the sensors used to collect real-time data can be prone to errors, said Professor Fernando Bernstein of Duke University’s Fuqua School of Business, as demonstrated by an experiment he conducted with Fuqua’s Bora Keskin at the at the Raleigh-Durham International Airport (RDU).

In research also co-authored by Adam Mersereau, Morgan Wood, and Serhan Ziya of UNC Chapel Hill, Bernstein and colleagues tested a system of people-counting sensors placed at the entrance and exit of the airport’s TSA security checkpoint. Their goal was to understand how busy the security area is at any given time, forecast waiting times, estimate the duration of the security check, and other metrics useful to optimize staffing and improve the passenger experience.

They found the sensor system they tested was highly unreliable, because it accumulated counting errors over time. However, they were able to fix the problem with an algorithm that produces estimates as close as possible to the actual number of people in the airport’s security area.

They also found that using inspections to manually corroborate the correct number of passengers significantly improved the estimates.

How managers monitor the usage of a service

Brick-and-mortar services like stores would need a lot of resources to track how many people are using their service, Bernstein said. They either need to physically inspect the traffic in a location — which is very expensive and time consuming — or they need to employ sensors and machines to automate the counting. 

In the airport experiment, Bernstein and colleagues placed infrared beam sensors at the entrance and exit to the TSA area. Every time a passenger breaks the beam, the system records an arrival or a departure. At any given time, the system records how many people have entered the area from the beginning of the day, and how many people have departed it. The difference between those numbers should be the number of people currently in the area.

The problem is that the sensors are inaccurate, Bernstein said.

“Somebody may be blocking the light beam, so there's no counting — sometimes for extended periods,” he said. “Or people may be walking side by side, and the system would count them as one, because they break the beam once. Or they have a carry-on luggage behind them, and the system counts them as two people.”

These errors accumulate over the course of the day, Bernstein added.

A fix to detect when the system may be empty

The researchers thought one way to improve the accuracy of the data would be to detect moments of the day when the TSA area was likely to be almost empty, and instruct the system to reset the count to zero, Bernstein said.

“If the net number the sensor is recording — arrivals minus departures from the area — is small enough, the system will reset the count to zero,” he said.

Resetting allows the system to stop accumulating errors, Bernstein added. 

The researchers compared the modified sensor data with the actual number of passengers transiting the TSA area and they found that their “resetting policy” produces a good estimate of the number of people in the system.

To make the estimate even more accurate, the researchers also introduced the possibility of using inspectors to count the passengers in the TSA area, allowing the system to reset to a real number.

“You could employ staff to periodically count the actual number of people in the area,” Bernstein said, “so it resets the counting problem.”

The researchers found the optimal level of manual counting that would improve the estimates at the lowest cost.

What’s the state-of-the-art on people counting technologies

People-counting systems use different technologies with varying costs, accuracy levels, and privacy issues, Bernstein said.

“Our infrared beam system is a pretty inexpensive technology,” he said. “It's completely anonymous, because it doesn’t use any camera.”

But there are other systems, potentially more accurate although more expensive, such as heat sensors mounted on ceilings.

“Think of a clothing store at the mall. If you wanted to install sensors mounted on the ceiling, you would need to drill and connect them with cables. Then you need to see what area each sensor covers. Let’s say the store is 100 square meters, so you may need 10 to 20 sensors and they need to be synchronized. This technology is more accurate, but it's a lot more expensive.”

Why tracking occupancy is relevant

Rapid advances in sensor technologies are offering a myriad of opportunities to collect real-time data on everything from machine performance, to produce freshness, and patient monitoring — they are also making it easier for managers to monitor office attendance

Similarly, public spaces like schools, churches and country clubs, which for safety reasons may limit the number of people who can be inside a facility, use these technologies to monitor occupancy levels, Bernstein said.

For an airport, sensor technologies are useful for staffing optimization and to improve the passenger experience, he said.

“Part of the experience of people flying through an airport has to do with the time they spend in line, and whether they risk missing a flight,” he said.

Collecting this data is important to forecast the use of the service — in a store, it’s part of the customer service to make sure you have enough staffing to serve the customers at any given time — and to understand customer behavior, Bernstein said.

“It's helpful for airports, for example, to estimate what they call a ‘show-up profile,’ which essentially is, ‘how early will you go to the airport?’ If you're going to New York for a business trip for a day, your show-up profile is going to be different from that of someone who is going with their family for a weekend,” he said.

“We are also collecting insights about passenger behavior, such as, ‘what type of passengers tend to arrive later?’ If your flight is early in the morning, or if you're going for business, you're more likely to arrive closer to flying time. So we try to not only provide a solution for forecasting, but also reveal insights as well.”