Turning your existing cameras into a business intelligence tool

Cameras that see a lot but tell you little

Most shops, restaurants, car washes, warehouses and industrial sites already have a CCTV system. It records around the clock, but the footage is only looked at after an incident. The rest of the time, those video streams produce nothing that helps you manage the business.

AI video analytics changes that. Instead of storing hours of footage nobody watches, it extracts events and indicators: footfall, how busy an area is, queue length, service time. Starting from what you already have makes obvious sense: the cabling, the cameras and often the recorder are already paid for and installed.

Not every camera suits every type of analysis, though. Before promising anyone a dashboard, a few prerequisites need checking.

Technical prerequisites to check

An audit of an existing installation usually looks at four sets of criteria:

• Stream access: recent IP cameras almost always provide an RTSP stream and often ONVIF support, which makes discovery and configuration easier. With a network video recorder (NVR), streams can sometimes be retrieved through the recorder itself. Older analogue cameras connected to a DVR are more restrictive: some DVRs expose a usable network stream, others do not.
• Resolution and image quality: there is no universal threshold. What matters is how large people or objects appear in the frame. A high-resolution camera placed too far away can be less useful than a modest one that is well framed. Compression, bitrate and frame rate also play a part.
• Positioning and angle: a camera installed for security usually points at a till, a door or a stockroom, not necessarily at the area you want to measure. Occlusion (shelving, pillars, crowds) and backlighting reduce reliability.
• Lighting and environment: day and night changes, reflections, shop windows, steam or water spray in a car wash, dust in a warehouse. Every setting has its own constraints.

Edge or cloud: where should the video be processed?

Once the stream is accessible, you need to decide where the analysis runs. There are two main approaches, which are sometimes combined.

Edge processing (Edge AI) runs on a small computer fitted with an AI accelerator, installed on site close to the cameras. Footage stays on the premises and only results (counts, events, durations) are sent to the dashboard. This approach greatly reduces bandwidth requirements, keeps working when the internet connection is unreliable and limits exposure of the footage, which is a real advantage for data protection.

Cloud processing sends the streams, or clips, to remote servers. It makes central maintenance easier but requires sufficient, stable upload bandwidth, which is not always available on Moroccan sites, particularly in multi-site networks or outside city centres. It also means the transfer and storage of footage must be carefully controlled.

At VIDEO.MA, we favour edge processing wherever possible and use the cloud for consolidating aggregated indicators and giving access to dashboards.

What you can realistically measure

With properly positioned cameras, the most common types of analysis are:

• counting people or vehicles at an entrance, across a line or within an area;
• occupancy of a space (dining room, terrace, car park, wash bay) over time;
• queue detection and estimated queue length;
• durations: time spent in an area, length of a service cycle or a wash;
• zone alerts: presence in a restricted area, a blocked emergency exit, prolonged stopping;
• consolidation of these indicators into dashboards by hour, day and site.

Combining video with your business data

A video indicator on its own tells you what is happening physically: how many people came in and how long they waited. It does not tell you what they bought. To link footfall with revenue, you need to integrate data from your point-of-sale (POS) system or ERP.

That integration depends on the systems in place and their export capabilities or APIs, and requires a dedicated study. A dashboard should always make a clear distinction between what is measured by video and what comes from a third-party system.

A few examples of useful comparisons: in a shop, the number of entries against the number of receipts; in a restaurant, dining room occupancy against the number of covers served; in a car wash, the number of vehicles detected against payment terminal transactions; in a warehouse, movements observed at a loading dock against the operations recorded in the management system. In each case, it is the gap between two sources that raises the right questions.

Start with a pilot

Rather than equipping an entire network from day one, we recommend a limited pilot on one site and a few cameras, with measurable objectives. It should run long enough to cover the usual variations in activity, weekdays and weekends, quiet and busy periods, so that the results are representative. A typical approach:

• define two or three precise business questions (for example: at what times does the queue exceed an acceptable length?);
• audit the cameras involved and, if needed, adjust the framing or add a better-placed camera;
• run the analysis over a representative period that includes quiet days and peaks;
• check reliability against manual counts on samples;
• decide, based on observed results, whether to extend, adjust or stop.

Limits to be aware of

Reusing existing cameras is not always the best option. Some are too old, badly positioned or of insufficient quality, and adding a camera dedicated to analytics can sometimes be more cost-effective. Accuracy is never perfect: it varies with crowd density, angle and lighting, and should be measured on site rather than assumed.

Finally, adding an analytics purpose to a CCTV system has data protection implications: informing people, completing formalities with the CNDP and setting retention periods. An approach built on aggregated indicators, processed on site without identifying anyone, keeps those issues to a minimum without making them disappear.

Want to know what your current cameras could measure? An audit of your installation is the best place to start.