Computer Vision for Restaurants: Hygiene and Waits

Every multi-branch operator in the Kingdom wants two answers on their desk each morning. Did every kitchen follow the hygiene rules yesterday? And how long did customers wait at the evening peak? Both answers are already on the cameras in your kitchens and at your counters. Computer vision turns that footage into a hygiene log you can show an inspector and a waiting-time number you can staff against.

This article is for operations directors, food safety managers and IT leads at restaurant groups, cloud kitchen operators and central kitchens in Riyadh, Jeddah, Dammam and beyond. It covers what the municipal requirements ask of your kitchen, what a camera can and cannot verify, how waiting times are measured, and what a pilot should prove.

What the municipal requirements ask of your kitchen

The Ministry of Municipalities and Housing publishes the Restaurant Requirements (1446 H), which apply to restaurants and cafés with and without service, cloud kitchens and food kiosks. A separate set of Central Kitchen Requirements (Arabic) covers central kitchens. The staff and food-handling rules in both are close to identical. Among them:

• Every worker must hold a health certificate.
• Food handlers must wear protective clothing and must not wear it outside the building or in the restroom.
• Workers preparing food must wear disposable, latex-free gloves, change them regularly, and change them after touching raw meat, unwashed produce or non-sanitised surfaces such as door handles.
• Masks must cover the nose and mouth in food processing areas.
• Hair and arms must be securely covered for workers preparing food.
• Hands must be washed with antibacterial soap at defined moments: before starting work, after a break, after the restroom, after handling raw food, and more.
• Deliveries must be checked on arrival, and chilled and frozen goods moved to temperature-controlled storage immediately. The requirements state that, as good practice, chilled food should not stay out of refrigeration for more than 20 minutes during transfer.
• The food preparation area must be visible to customers or separated by transparent glass, with the option of using cameras and screens.
• Security cameras must be installed in line with the Law of Using Security Surveillance Cameras and its Implementing Regulation.

These rules are checked at inspection. Between inspections, most operators rely on a supervisor's paper checklist. That is the gap computer vision closes.

What a camera can verify, and what it cannot

Honesty first. Vision is strong on visible behaviour and timing. It is weak on anything that needs a document, a thermometer or a close look at a hand. A good programme uses each tool for what it does best.

Requirement	What computer vision can check	What it cannot replace
Masks in processing areas	Mask on or off, per station, per hour	Mask quality or fit
Hair and arm cover	Hair net and sleeves present at preparation stations	Whether the cover is clean
Gloves	Gloves worn at preparation stations	Every glove change, unless a camera covers the station closely
Handwashing	Visits to the handwash sink at the start of shift and after breaks, with duration	Whether soap was used and hands were dried properly
Protective clothing outside the kitchen	Kitchen uniform seen in restroom corridors or at the exit	Anything inside the restroom; cameras are prohibited there
Deliveries and cold chain	Time from delivery arrival to goods entering the cold store	The temperature itself, which needs a probe or sensor
One-direction flow	People moving from raw preparation to ready-to-eat areas without passing the sink	Cross-contamination that is not visible
Health certificates	Nothing	That is a records check

The result is not a camera that replaces your food safety manager. It is a manager who starts the day with a list of the twelve moments that need attention, each with a short clip, instead of a stack of ticked boxes.

From observations to a hygiene log

A hygiene log is useful only if someone reads it and acts. Build it around stations and shifts, not around named individuals.

• One event per observation. Time, branch, station, rule and a short clip with faces blurred.
• A daily summary per branch. Observations by rule and by shift, compared with the previous week.
• Acknowledgement. The branch supervisor marks each event reviewed, coached or disputed. Disputed events go back into the accuracy review.
• Trend, not blame. Repeated misses at one station usually point to a layout or supply problem: the glove box is on the wrong wall, the sink is too far from the cutting table. Fix the station first.
• Evidence when you need it. When an inspector or a franchisor asks how you monitor compliance, the log shows dates, counts and actions.

Events can flow into an HSE register, a quality management system or a BI dashboard, so the log sits next to the rest of your food safety records rather than in another app.

Waiting times: the front-of-house numbers

Waiting time is the number customers feel and few restaurants measure. Order timestamps in the POS show when an order was placed, not how long the customer stood in line before placing it.

Cameras over the counter, the pickup area and the drive-through lane fill that gap:

• Queue length and time to order. How many people are waiting and how long the line takes to reach the cashier, minute by minute.
• Order to handover. Matched with POS timestamps, the time from order to food in hand.
• Walk-aways. People who join the queue and leave before ordering. This is lost revenue that never appears in sales data.
• Drive-through flow. Vehicles in the lane, time per vehicle and the point where the lane backs onto the street.
• Dining-area occupancy. Tables in use by hour, for staffing and for planning seating at new branches.

For cloud kitchens, the Restaurant Requirements also oblige management to regulate delivery riders' movement to prevent congestion inside and outside during order pickup and handover. A camera on the handover area measures rider count, rider dwell time and the minutes when the area overflows. Those minutes tell you which brands in the kitchen need a pickup slot system, and which hours need an extra runner.

Peaks in the Kingdom are predictable but steep: Ramadan evenings, weekends, paydays, school holidays and major events. A waiting-time baseline from a normal week lets you plan staffing for the peak instead of reacting to it.

Central kitchens: a different problem

A central kitchen is a small factory. The questions shift from the counter to the line.

• Receiving. Time from truck arrival to goods in the cold store, against the 20-minute good-practice limit in the requirements.
• Line throughput. Trays or batches per hour per line, and idle time between batches.
• Zone discipline. Movement between raw and cooked zones, and visits to handwash stations at zone changes.
• Dispatch. Time from packed to loaded, and vehicles waiting at the dispatch dock.

Our guide on what your existing CCTV cameras can already tell you covers dock and yard measures in more depth.

Staff are data subjects too

Kitchen footage shows people. The Personal Data Protection Law defines personal data to include photos and videos of an individual, and the principles published by SDAIA require purpose limitation, data minimisation and destruction once the purpose is fulfilled.

Restaurants and other food service establishments are also listed among the categories covered by the Law of Using Security Surveillance Cameras (official Arabic text), which prohibits cameras inside changing rooms and toilets. Its Implementing Regulation allows entities that are not otherwise authorised to analyse security camera recordings only where people are not identified or disclosed, and only after approval from the Public Security Directorate. Confirm with your legal team how this applies to each camera you plan to use.

TQN designs for this from the first camera. Models run on edge hardware in the branch or the central kitchen. Only events and short clips leave the site. Faces are blurred by default, and there is no face matching unless you request it and it is lawful. Hosting is on your premises or in a Saudi data centre. The system is designed to support PDPL and NCA requirements.

What a pilot proves

A pilot runs four to six weeks on your own footage, in one or two branches or one central kitchen, against targets agreed before it starts. You should leave with:

• Accuracy per rule. For each hygiene rule, how often an alert was correct (precision) and how many real misses it caught (recall), against a manually reviewed sample. The method is the same one we describe for PPE detection on industrial sites.
• False alarms per camera per day. If supervisors receive noise, they stop reading.
• Waiting-time accuracy. System measures compared with a stopwatch sample taken by your own team at peak hours.
• A baseline. Today's hygiene observations, queue times and walk-aways, so the next quarter has a starting point.
• A rollout view. Which branch camera layouts work as installed and which need an angle change.

The pilot scope and price are fixed and agreed before it starts. The assessment before it is free. For more on how our Vision AI practice works across food and beverage, see computer vision for restaurants and cafés.

Show us one camera

Choose one camera: the one over your busiest preparation station, your counter or your cloud kitchen handover area. Send us ten minutes of footage. We will send back a free read-out of what it could report, which hygiene rules it can see, which waiting-time measures it supports, and what limits its angle or light.

Show us one camera and see your kitchen the way an inspector would, every day.