Queue detection: from alert threshold to operational decision

Measuring a queue: what a camera can estimate

A queue that gets too long has a cost: customers who give up, frustration, tension at the counter. Yet in a shop, a bank branch, a clinic or a public service, it is often noticed too late. Camera-based queue detection aims to put numbers on it.

In practice, a zone is drawn in the image to match the waiting area, in front of a till, a counter or a reception desk. The system detects the people in that zone and produces two kinds of indicators:

• the number of people queuing, measured continuously;
• an estimated waiting time, calculated from time spent in the zone or from the rate at which people leave the queue;
• derived indicators: how long the queue stays above a threshold, the longest queue per time slot, the number of saturation episodes.

Setting thresholds that mean something

An alert is only worth sending if it leads to action. So the threshold should start from an operational question: at how many people, or how many minutes of waiting, should another till or counter open? The answer depends on the business, service capacity and customer expectations.

A few principles help set sensible thresholds:

• combine a count and a duration: more than five people for more than three minutes, rather than more than five people at any given moment, which avoids alerts on a brief spike;
• tailor thresholds to the time slot or site: the same queue length means something different on a Saturday lunchtime and a Tuesday morning;
• add a cooldown before re-alerting on the same queue, giving the response time to take effect;
• start with conservative thresholds during the pilot, then adjust them based on the alerts that proved genuinely useful.

Avoiding alert fatigue

The biggest risk with an alerting system is not missing an alert, but sending too many. When a manager receives notification after notification, often with nothing they can do about them, they end up ignoring them all, including the ones that matter.

During the pilot, it pays to track the number of alerts per day, the share that led to an action and the false alerts caused by specific situations: people standing around without queuing, staff in the zone, a delivery. These observations are used to refine the zone, the thresholds and the active hours. A good set-up aims for few alerts, but relevant ones.

Alert channels and response procedures

An alert has to reach the right person through the right channel: a dashboard notification, a message on a duty phone, a display at the supervisor’s desk. The choice depends on how the site is organised and who is available.

Above all, the response must be agreed in advance. Who opens the extra till? Who steps in at the counter? What happens if nobody is free? A simple written procedure that the team knows turns a notification into a decision. Without one, the alert is just another piece of information.

It also helps to close the loop: record whether the alert was picked up, what was done and how long the queue took to clear. Even brief feedback like this shows whether thresholds are well set and whether the organisation keeps up. Over the weeks, it also reveals whether the problem is solved by one-off reactions or calls for a more structural decision, such as an extra staff member on a recurring time slot.

Queue alerts are service management tools. They never replace certified safety systems such as fire detection or evacuation management.

Reading the trends: hours, days, sites

Beyond real-time alerts, queue history often delivers more lasting value. By consolidating measurements over several weeks, you can pinpoint time slots that are consistently overloaded, compare days of the week, see the effect of an organisational change and compare sites measured the same way.

These trends feed into staff rotas, the number of open tills or counters and the layout of the waiting area. Combined with data from other systems, such as till receipts or a queue management ticketing system, they can go further, provided the integration has been studied and each source remains clearly labelled on the dashboard.

One word of caution: a video-based waiting time is an estimate, not an individual stopwatch. It is reliable as a trend, for comparing time slots or sites measured the same way, but it should be validated against manual measurements before being used as a service commitment displayed to customers.

Limits worth knowing

Queue detection works well when the waiting area is structured and clearly visible. It gets harder in certain situations:

• loose queues with no barriers, where people wait scattered around or seated;
• obstructions from displays, pillars or dense crowds;
• queues that merge or cross, for example in front of two neighbouring tills;
• people in the zone who are not actually queuing: companions, staff, passers-by.

From measurement to decision

These limits do not make the tool useless, but they do mean the measurements must be validated on site, with manual checks, before you rely on them. The analysis covers flows, not individuals: no identification is needed, and the data controller must inform people and complete any applicable CNDP formalities under Law 09-08.

Looking to manage waiting times better in your stores, branches or healthcare facilities? Explore our solutions and industry pages, then request an audit to prepare a pilot with thresholds that fit your operations.