Managing wait times at an airport requires measuring them where they occur—zone by zone and time slot by time slot—rather than on a platform-wide basis. An average wait time of eight minutes at security checkpoints tells us nothing: it could mask a line that moves smoothly all day and two 40-minute peaks in the morning—which are precisely the times when operations need to take action.
Useful measurement relies on three metrics calculated continuously by zone: the number of passengers in the line, the throughput at open checkpoints, and the resulting estimated wait time. These three metrics are derived from cameras already in place, without identifying any individuals. XXII, a French computer vision software provider, generates these types of indicators using its CORE platform, without facial recognition.
An airport complex comprises several checkpoints with different constraints: baggage drop-off, security screening, border control, boarding, and sometimes a dedicated security checkpoint for connecting flights. Each has its own capacity, its own operating hours, and its own sensitivity to arrival waves.
Aggregating these checkpoints into a single indicator masks where the problem actually lies. A growing security screening line, while border control remains smooth, does not call for the same decision as widespread congestion. Dividing the area into zones is therefore essential for the analysis to lead to concrete action.
The breakdown is based on possible decisions, not on the building’s layout. A zone must correspond to an area where someone can take action: open a checkpoint, redirect a flow of passengers, call for backup, or inform passengers.
Airport traffic is structurally irregular. Departure waves are concentrated in a few time slots; the distribution varies by day of the week; and the season shifts the entire curve.
Breaking down traffic by time slot—typically fifteen or thirty minutes—enables three types of analysis that a daily average cannot provide. Comparing the same time slot from one day to the next, which isolates the effect of an operational decision. It allows for the detection of a discrepancy between the expected wave and the actual wave, in the event that a group of flights is delayed or ahead of schedule. Finally, it enables the calculation of actual capacity utilized per time slot, which serves as the basis for staffing levels.
This time step has another advantage: it corresponds to the operational horizon. Decisions to open an additional inspection station, call in a backup agent, or modify a display are made and carried out within this timeframe.
This is measured by counting within a polygon corresponding to the waiting area. It is the most reliable raw data and the least subject to interpretation.
The number of passengers processed per minute per station indicates the system’s actual capacity at that moment. When compared to the number of open stations, it reveals performance differences between various configurations.
It is calculated based on the ratio of the line length to the observed throughput. This estimate is more robust than a measurement of individual travel times and does not require any tracking of specific individuals. However, it must be regularly recalibrated based on field observations to remain credible with the teams.
It indicates the level of comfort and the available margin before the line spills over into traffic lanes—a situation that rapidly degrades the entire sector.
Four decisions come up regularly in operations centers.
Opening and closing security checkpoints. A threshold defined by zone, combined with the expected passenger volume for the next time slot, allows for the opening to be anticipated rather than triggered only after the line has formed.
Redirecting some passengers to a less congested checkpoint, when the airport layout permits it. This decision requires clear signage placed before the choice point.
Deploying reception agents to areas where passenger density is rising, rather than following a fixed schedule established at the start of the day.
Finally, keeping passengers informed. Displaying the actual wait time—even if it’s long—reduces frustration and limits the number of inquiries directed at staff. An inaccurate estimate has the opposite effect, making regular updates essential.
The value of continuous measurement lies less in the instantaneous figure than in comparison. Three comparisons are particularly useful.
Comparing a time slot to its equivalent from previous weeks highlights a gradual drift, often linked to changes in the flight schedule or procedural adjustments.
Comparing zones with comparable traffic highlights performance discrepancies. Two checkpoints handling similar volumes but with different throughput rates indicate differences in equipment configuration, organizational structure, or passenger preparation.
Finally, comparing measured wait times with stated theoretical capacity informs investment decisions. It helps distinguish between a structural bottleneck—which requires physical modifications—and a scheduling issue—which calls for a different organizational approach.
Three factors determine feasibility for an existing camera system.
The field of view and resolution must allow for the identification of individuals throughout the entire waiting area, even when the line extends beyond its usual footprint. Cameras installed for verification purposes are often mounted too high to provide reliable counting at the back of the area.
Access to the video streams must be possible via the local network or the recorder, with the appropriate permissions and without requiring major reconfiguration.
The processing method must account for latency constraints and service continuity. On-site analysis continues to function even during a connection outage and transmits only the calculated data, which significantly reduces the required bandwidth.
The GDPR applies whenever processing involves personal data, and the image of an identifiable person falls under this category. Measuring a line length and throughput does not require any facial templates. A solution that does not use facial recognition avoids the biometric data regime set forth in Article 9 of the GDPR, which is reserved for special categories of data.
Data minimization reinforces this position: the system may produce only aggregated counts and durations, without retaining the images used for the calculation. The standard obligations remain in effect: informing individuals via posted notices, a clearly defined and non-misappropriated purpose, a limited retention period, and an impact assessment when processing warrants it. The CNIL has published a position paper on so-called “augmented” cameras, which is useful for defining the intended purpose. When employees are involved—particularly inspection staff whose throughput is measured—consultation with employee representative bodies is required under the conditions set forth in the Labor Code, and the purpose must remain the management of workflow rather than individual evaluation.
Select an area and a decision. A screening checkpoint, a threshold for opening a lane, or a wait-time display are sufficient to establish an initial complete loop: measurement, threshold, action, and verification of the effect. Adjust the estimate based on field data collected during the first few weeks, then expand to other areas using the same framework; otherwise, comparisons between areas lose their value.
XXII’s CORE platform was designed for this type of phased rollout across an existing camera network, with indicators by zone and time slot that can be exported to existing monitoring tools.