Implementing a system to measure airport security line lengths using existing cameras requires four elements: access to the camera feeds covering the waiting area; the division of this area into segments (line entry, winding area, security checkpoints), a model that counts people and estimates the travel time between the line’s entry and exit points, and the distribution of this metric to the relevant stakeholders—namely, operations management, passenger information displays, and staffing planning.

The challenge of the project is not the calculation; it is the integration. An accurate measurement that does not appear on passenger screens or in the planning tool has no impact on operations. This article describes the stages of a deployment and the factors that determine its success. XXII, a French computer vision software provider, handles this type of scenario using its CORE platform, which does not use facial recognition.

Which metrics should be generated, and for which recipients?

Three metrics cover most needs, and each is intended for a different audience.

The estimated wait time, expressed in minutes, is the metric intended for passengers and information screens. It must be stable and conservative: an estimate that fluctuates every two minutes undermines the display’s credibility.

The number of people in line and the hourly throughput per station are the metrics intended for management. They enable decisions to be made about opening an additional station before the line becomes overcrowded, rather than after.

Historical data broken down by time slot, day, and season is the metric intended for planning. It is used to determine staffing levels during peak periods and to measure the impact of organizational changes.

Defining these three uses before deployment avoids the most common pitfall: producing accurate data that no one uses because it has not been linked to a decision.

Which cameras are suitable, and which are not?

The monitoring area is generally well-covered by cameras, but these cameras were installed for security purposes, with angles designed for scene recognition rather than for counting.

A frame suitable for measuring a line must meet three conditions. The camera must see the entire measured segment, or, failing that, clear entry and exit points. The angle must be sufficiently downward to minimize obstruction by other people, a phenomenon that skews counts in dense lines. Lighting must remain stable, which can sometimes be an issue under a glass canopy at the end of the day.

When a measurement point is missing, two options exist: add a camera at that specific point, or adjust the zone’s boundaries to measure a slightly different segment. The second option, which is less expensive, is sufficient in many cases.

Where should data processing take place—on-site or at a central facility?

An airport generally has a robust network infrastructure, which leaves both options open. The choice is based on three criteria.

Bandwidth. Streaming video feeds continuously to a central processing facility consumes significantly more bandwidth than local processing, which transmits only counter data.

Security policy. Many operators do not allow video streams to leave the restricted area. Processing performed within the perimeter, with transmission limited to non-identifying data, simplifies the review of the case.

Service continuity. Local processing continues to generate measurement data in the event of a connection outage, with the data subsequently transmitted on a delayed basis.

How can meter readings be integrated into existing systems?

This step determines the value of the project. Three integrations must be planned from the design phase.

Passenger displays, which require an exchange format compatible with the existing information system and clear rules regarding refresh frequency and the displayed rounding.

Monitoring, which requires alert thresholds defined in collaboration with operations teams—for example, an estimated wait time exceeding a target value for a given duration, rather than an alert triggered by every momentary exceedance.

Scheduling, which requires historical data exports in a format compatible with workforce management tools. An API or regular data export is generally sufficient.

Ask each candidate for the exact format of the data produced and the documentation for its interface before signing a contract. This is a difficult issue to correct after the fact.

What is the legal framework for this type of measure?

The GDPR applies whenever processing involves personal data, and an image of an identifiable person falls under this category. Two principles allow for a straightforward approach.

The first is the absence of biometrics. Measuring a line involves counting people and estimating wait times, not identifying them. Facial recognition falls under biometric data, which is governed by Article 9 of the GDPR as a special category of data, with a general prohibition accompanied by limited exceptions. A crowd-counting measure that does not use facial recognition avoids this regime.

The second is data minimization: producing aggregated counts and durations without retaining the images used to calculate them. The usual obligations remain: informing individuals, specifying the purpose, limiting retention periods, and conducting an impact assessment when the processing warrants it. The CNIL has published a position paper on so-called “augmented” cameras that helps define their use. This topic should be addressed with the data protection officer as early as the scoping phase, rather than at the time of deployment.

What steps should be planned for deployment?

Deployment is structured into four phases.

Scoping: selection of indicators, recipients, and thresholds; identification of suitable cameras; and validation of the processing plan with the security and IT teams.

A pilot at a checkpoint, including a phase to compare the data generated with manual readings taken during several representative time slots, including peak hours. It is this comparison that establishes the indicator’s credibility with the teams.

Integration into the three target systems, starting with supervision, then scheduling, and finally passenger displays—which are the most visible and should be activated once the measurement has stabilized.

Extension to other checkpoints, while maintaining the same indicator definitions. A definition of wait time that varies from one terminal to another makes any comparison impossible.

What CORE brings to this use case

XXII’s CORE platform analyzes footage from existing cameras and generates real-time operational metrics, including people counting, area occupancy, and dwell times—all without facial recognition. At an airport, the benefit lies in the ability to gradually cover multiple checkpoints using the same indicator definitions and the same monitoring system, leveraging the existing camera infrastructure rather than installing dedicated sensors at each location.