At a gas station, AI-powered video analysis is primarily used to detect four types of incidents in real time: falls and fainting incidents on the fueling island or in the store, unauthorized access to technical areas outside of business hours, vehicles driving in the wrong direction or parking dangerously, and suspicious behavior around electric vehicle charging stations. In each case, the software analyzes the footage from existing cameras and sends an alert to the operator or the monitoring center.

The difference from conventional motion detection lies in how the scene is interpreted. A motion detector signals any change in the image, including a shadow or a curtain of rain. A computer vision model distinguishes a person from a vehicle, measures the duration of a presence, and recognizes a person’s stance or the direction of traffic flow. This is what makes it possible to reduce the number of unnecessary alerts to a level compatible with real-world operations. XXII is developing this approach in France with its CORE platform, which does not use facial recognition.

What security incidents does video analysis detect on a runway?

The runway poses the most immediate risks because people and vehicles share the same space, often without any physical separation.

Fall detection identifies a person who has fallen and remains on the ground for more than a few seconds. At a site operating extended hours or in unmanned, automated mode, this alert triggers a call before a third party reports the situation.

Detection of wrong-way driving and abnormal traffic patterns relies on comparing the observed trajectory with the expected direction of travel in the area. It flags a vehicle driving against traffic, cutting across an exit lane, or parking in front of a fire department access point.

Detection of prolonged presence flags a vehicle that has been stationary at a pump for significantly longer than the time required to fill up, which can indicate either a breakdown or unauthorized parking blocking a pump during peak hours.

How can intrusions be detected outside of business hours?

Technical areas, storage rooms, car wash facilities, and the areas around fuel tanks are spaces that should be completely unoccupied at night. These are therefore areas where a simple rule works well: any human presence detected within a defined perimeter during a given time period triggers an alert.

Video analysis provides a direct benefit here. By distinguishing a person from an animal, a plastic bag, or a car headlight passing on the nearby road, it prevents the majority of false alarms that ultimately lead to the system being ignored. An alert accompanied by the corresponding image also allows the operator to resolve any uncertainty without having to go to the site.

The same logic applies to areas that are off-limits during the day, such as the vicinity of a fuel tank during a delivery or a temporarily marked construction zone.

What should be monitored around electric charging stations?

Charging stations present challenges that traditional fueling stations do not. Parking durations are measured in tens of minutes, the equipment is exposed, and the cables contain copper.

Three types of detection are commonly implemented: the presence of a person near the stations outside of a plausible charging session, particularly at night; prolonged handling of a cable or connection point, which distinguishes a normal connection from an abnormal intervention; Finally, a charging station being occupied by a vehicle that is not charging—an issue more related to operations than security, but one that impacts service availability.

How can the customer journey be tracked using the same cameras?

Cameras installed for security also generate operational data, provided the camera angles allow for it. At a gas station, the most commonly used metrics include counting vehicles entering the service lanes, counting customers entering the store, measuring checkout line wait times, and tracking service lane occupancy rates by time slot.

These metrics inform concrete decisions: staffing teams during peak times, revising signage that fails to direct drivers to the store, or identifying times when checkout wait times result in lost sales. The benefit of combining security and operations using the same camera system is clear in terms of cost, provided that the camera angles are suitable for both purposes.

How can you compare the available solutions?

Five criteria help distinguish between offerings that, on paper, provide the same detection capabilities.

Compatibility with the existing infrastructure. A solution that requires replacing cameras completely changes the economic equation for a network of stations.

The false alarm rate under real-world conditions, measured at your site rather than during a demonstration. A rainy night, partial lighting, and nearby traffic constitute the true test.

Processing method. On-site processing reduces the required bandwidth and allows the system to continue operating in the event of a network outage, which is crucial for remote sites.

Multi-site monitoring. When there are more than a few stations, you need to be able to verify from a single location that each camera is transmitting properly and deploy updates without local intervention.

Compliance. The GDPR applies whenever processing involves personal data. A solution that does not use facial recognition avoids the biometric data regime set forth in Article 9 of the GDPR and simplifies the impact assessment. The CNIL has published a position paper on so-called “augmented” cameras that helps define the intended purpose.

What obligations must be met with regard to customers and staff?

Three obligations form the framework of the project. Informing individuals through a visible sign at the site entrance indicating the existence of the system, its purpose, and the means to exercise their rights. Purpose limitation, which prohibits the reuse of a system installed for security purposes for the individual monitoring of staff. Retention period limitation, which must be defined and documented for each type of data generated.

When the system involves employees, consultation with employee representative bodies is required under the conditions set forth in the Labor Code. A project presented in advance, with a precise description of what is being measured and what is not, generally goes more smoothly than a system discovered after it has been installed.

Where to start with a network of stations?

Select two or three priority detection criteria—those whose absence is currently the most costly. Test the system at a site representative of the network’s most common constraints, under both day and night conditions, for a sufficient period to assess the false alarm rate. Then expand in phases, maintaining the same measurement grid. XXII’s CORE platform was designed for this type of rollout on an existing camera fleet.