A camera system installed for security purposes can provide operational metrics without replacing the equipment, provided that three parameters are compatible with the intended use case: the effective resolution over the measured area, the frame rate, and the field of view. When one of these three parameters is unsuitable, the solution is generally to reposition a camera or add a single targeted camera, not to replace the entire system.

The distinction to keep in mind is simple. A security camera is installed so that a human operator can understand a scene after the fact. An analytics camera is installed so that an algorithm can produce a repeatable measurement. The two uses often overlap, but not always, and it is this overlap that must be evaluated on a site-by-site basis before committing to a project.

What an Existing Park Allows Us to Measure

Several categories of metrics can be obtained using traditional security camera views—that is, wide-angle views positioned at a height.

Occupancy and availability: occupied dock spaces, available storage locations, saturated preparation areas, and the duration for which an area remains congested.

Traffic flows: number of material handling vehicle passes on an aisle, direction of traffic, transit time between two zones, truck queues at the site entrance.

Compliance with operating rules: presence in a restricted area, a pedestrian walking on a lane reserved for vehicles, a loading dock door left open, prolonged parking in front of an exit.

Time-related metrics: duration of a truck’s stop at a loading dock; time between a vehicle’s arrival and the start of unloading; cycle time of a repetitive operation when the area is properly framed.

These metrics have one thing in common: they pertain to large objects, areas, and durations. This is precisely what a security camera can capture.

What a security camera cannot do

The limitations are just as clear, and it’s best to be aware of them before promising a metric to operations management.

Reading fine details from a distance. Reading a serial number, a code on a label, or a floor marking from a camera positioned eight meters high requires a level of resolution that most wide-angle views lack. The key factor is the number of pixels available on the object itself, not the sensor’s nominal resolution.

Rapid movements. A low frame rate—common in older systems configured to conserve storage—is sufficient for measuring occupancy but not for characterizing brief movements. Gesture recognition or machine motion analysis requires a significantly higher frame rate.

Heavily Obstructed Scenes. Dense shelving, a partially obscured conveyor belt, or a cross-flow of vehicles create permanent obstructions. The algorithm can compensate for occasional obstructions, but not for an area that is only partially visible.

Extreme lighting conditions. Midday backlighting from a loading dock door, a wash area, highly variable lighting, or an infrared zone at night degrade measurement accuracy. These situations do not preclude the use of the system; they simply require on-site validation.

How do you assess a fleet before launching a project?

The preliminary audit consists of four steps and rarely takes more than a few days.

List the cameras, including their model, resolution, configured frame rate, and location. This list is often already available from the integrator responsible for maintenance.

Associate each desired indicator with the camera(s) that cover the relevant area. Many projects stop here, upon discovering that no camera actually covers the area of operational interest.

For each selected camera, extract a real-world sequence covering a full day, including a nighttime period and a period of high activity. An evaluation based on a ten-minute demonstration clip has no predictive value.

Finally, define—indicator by indicator—the acceptable error margin. A vehicle count with an error margin of a few percent may be sufficient for a weekly trend but insufficient to trigger a real-time alert.

Should images be processed on-site or remotely?

Both architectures exist, and the choice depends primarily on the network connection and the company’s internal policy.

Processing performed on equipment installed on-site limits the volume of images leaving the premises: only the measurements—and sometimes a few thumbnails associated with alerts—are sent to the management tools. This option is suitable for sites with limited connectivity, which is still common at logistics hubs and isolated production sites.

Remote processing simplifies IT infrastructure management but requires constantly available bandwidth and a clear policy on the location of processing and data.

What obligations apply when employees are filmed?

At an industrial site, cameras record employees and external contractors. The project is therefore subject to the GDPR, regardless of whether the metrics produced are aggregated.

Three obligations form the basis of the requirements: providing individual notice to data subjects via a written notice and signage in the covered areas; Consultation with the social and economic committee prior to implementation, as the surveillance systems fall within its purview. And, in most industrial settings, a data protection impact assessment, given the continuous monitoring of work areas.

One principle guides the rest of the process: the monitoring must focus on areas and workflows, not on identified individuals. Formalizing in writing that the metrics are not used to evaluate individual employees facilitates discussions with employee representatives and limits legal risk.

What XXII Offers

XXII is a French computer vision software developer. Its CORE platform is designed to leverage existing cameras and operates without facial recognition. It generates metrics on areas, traffic flows, and duration that can be used by operations teams, and can operate with local processing when the site’s network connection is limited.

What’s the best first step?

The best starting point is a single metric—one that a team is already expecting—for an area that is adequately covered. Loading dock occupancy and truck dwell time are good candidates: the area is unobstructed, the objects are large, and the measurement has an immediate recipient.

Once this metric is reliable and in use, expanding to other areas can be discussed based on actual data. It’s the opposite approach—the one where a project aims for ten metrics from day one—that most often results in a dashboard that no one looks at.