Case study · Our own product
VisionOps
A till knows what was sold. It doesn’t know what happened on the floor: how long a table waited for food, which shelf people stopped at, how long the queue was, or which seats sat empty. VisionOps measures that with one overhead camera and a small computer on site.
- Type
- Our own product, not client work
- Where it fits
- Restaurants, shops and any space where people sit, queue or browse
- Built
- Edge vision engine, zone studio, dashboard
- Status
- Trial built; first venue install under way
See it run
An airport gate, 24 seats, every one timed.
The same engine, set up for a waiting area: a zone per seat, people tracked as they walk through, and a timer on every seat that’s in use. Swap the seats for tables, displays or a till queue and it works the same way in a restaurant or a shop.
The brief
See the service, not just the sales.
The questions owners ask on a busy day are about the floor. In a restaurant: which tables are turning, and how long food waits before it reaches the table. In a shop: where people stop, and how long they queue. Anywhere: how many people came through, and how long each space was in use.
We wanted to answer them with hardware a small business can afford, without filming faces or sending video anywhere, and with every number traceable back to the moment that produced it.
Where it works
One engine, any floor with people on it.
The camera, tracking and timers are the same everywhere. What changes is the zones you draw and the questions the dashboard answers.
Restaurants and bars
Zones for each table, the kitchen pass and the bar rail. Each table runs through its own states, from seated to served to cleared.
- Table turns and how long each sitting lasts
- Kitchen-to-table time
- Minutes lost to tables nobody has cleared
Shops
Zones for the door, each display or aisle, and the till queue, linked to sales once the till is connected.
- Footfall through the door, hour by hour
- Which displays people stop at, and for how long
- Queue length and waiting time at the till
Anywhere people wait or gather
Waiting areas, gates, receptions, offices and event spaces. Draw a zone per seat, desk or area and the same timers apply.
- Which seats or desks are in use, and for how long
- How many people passed through, not just sat down
- Busy and quiet times across the day
What we built
Camera to dashboard, without the video leaving the room.
The trial version works from uploaded video rather than a live camera, so every test can be run again and give the same answer. Everything else is the full system.
One camera, looking straight down
A single overhead IP camera covers the floor: seating, aisles, counters and doorways. Looking straight down avoids people blocking each other and gives a clean head-and-shoulders point to track.
Detection and tracking on site
A small Intel computer on site reads a low-resolution feed at five frames a second, detects people with YOLO11n and follows each one with ByteTrack, accelerated with OpenVINO.
A zone studio for any floor plan
Pause on a frame and draw the zones by hand: a table, a seat, a display, a queue, a till or a doorway. Layouts can be saved and reloaded.
A state machine for every zone
Each zone has states and time rules. A restaurant table moves between vacant, occupied, service visit, needs busing and turned over; a seat or display is simply free or in use. People just walking past are ignored.
Playback and a dashboard
A video player draws the tracks and coloured zones over the footage. The dashboard shows how long each zone was in use, how many people came through, and in a restaurant, tables turned, kitchen-to-table time and time lost to unbussed tables. Every event is tied to its moment in the video.
Private by design
No facial recognition. The video is analysed on site and never uploaded; only small event messages leave the building, over an outbound encrypted connection.
In a restaurant: how a table changes state
Four rules, each with a reason
Short timers stop the system reacting to every passing body, so a table only changes state when something has really happened.
Vacant → Occupied
Someone stays in the table zone for 10 seconds
People walking past are ignored
Occupied → Service visit
A person who came from the kitchen pass stays at the table for 5 seconds
Logs when food or drinks arrived
Occupied → Needs busing
The table has been empty for 30 seconds
Starts the clock on lost table time
Needs busing → Vacant
Staff stay at the table for 15 seconds
Marks the table as turned over
| Change | When | Why |
|---|---|---|
| Vacant → Occupied | Someone stays in the table zone for 10 seconds | People walking past are ignored |
| Occupied → Service visit | A person who came from the kitchen pass stays at the table for 5 seconds | Logs when food or drinks arrived |
| Occupied → Needs busing | The table has been empty for 30 seconds | Starts the clock on lost table time |
| Needs busing → Vacant | Staff stay at the table for 15 seconds | Marks the table as turned over |
The trial
Five moments every service has
The trial is built to be checked against five scenes, each with a clear right answer. We’ve been running it on recorded restaurant and CCTV footage, comparing several head-detection models for the overhead view.
- 01
Someone walks past a table
The table stays vacant.
- 02
Two people sit down for a meal
The table turns occupied after 10 seconds and the dwell clock runs.
- 03
A server brings food from the pass
A service visit is logged against that server’s track.
- 04
The guests leave
A needs-busing alert fires 30 seconds after the table empties.
- 05
Staff clear the table
The table resets to vacant and the lost time is recorded.
Where it stands
Trial built. First venue next.
The vision-only trial is built and running on recorded footage. A first live installation, in a bar, is under way. What comes next:
- Joining the till: matching each table’s session to its tickets, for revenue per seat-hour and order-to-delivery wait.
- Bar versus table turnover, and how quickly each member of staff gets orders out.
- Fusing two camera feeds where one overhead view can’t cover the room.
We’ll add results here once the first venue is live and has agreed to share them.
For your venue
Run a restaurant, shop or venue?
Footfall and in-store behaviour
On-site cameras count visitors and show where they spend time, linked to your sales so you can see what converts.
Shelf and stockroom cameras
Cameras count stock and spot empty shelves, feeding live levels into your dashboard.
Dashboards
Custom dashboards built around the questions you ask: a Looker Studio command centre for you, plus live screens for the office or warehouse showing today’s orders, stock alerts and targets.