How AutoVision measures pump occupancy time
Operators should be able to check how a number was made. This page documents what AutoVision measures, the rules for a valid pump session, how averages are calculated, and the limits we know about.
What is measured
AutoVision measures pump occupancy time: the time between a vehicle entering a pump bay and leaving it, as seen by the forecourt cameras. It does not detect dispensing, so it never reports fuelling time.
What counts as a session
- Both an entry and an exit must be observed. A vehicle still on the pad is in progress and is not averaged; it is counted separately.
- Short gaps in tracking are merged: the same tracked vehicle, or the same plate within a short window, is one session.
- Very short detections (under 15 seconds for plated vehicles, under 20 seconds for unplated ones) are treated as passing traffic and excluded. These are configurable.
- Sessions above 30 minutes are excluded as implausible.
- Two sessions can never overlap on the same pump, and a session is never counted twice.
How the average is calculated
The average is the total occupancy time of all completed sessions divided by the number of those sessions, so a busy pump counts for more than a quiet one. It is not an average of per-pump averages. Seconds are shown to one decimal place and minutes are derived from the same unrounded value.
Every figure is calculated once and shown identically on the dashboard, the live view, the daily and executive reports and the API for the same period. Each average is shown with its sample size.
What we deliberately do not do
- We do not hide short sessions to make an average look longer. We publish what the average would be under stricter rules, for transparency, but the headline uses the stated rules.
- We do not call occupancy time fuelling time.
- We do not show a make or model for a vehicle unless the system is confident; less certain guesses are labelled as likely.
Known limits
Tracking can occasionally split one vehicle into two consecutive sessions, which makes the average slightly shorter and the session count slightly higher. Number plates misread across frames can make one vehicle look like two. We cannot always tell these apart from a genuine quick changeover, so we report the effect as a bound instead of correcting it silently.
Occupancy also includes the time a vehicle stays after fuelling, such as paying or waiting for a passenger.
How we check our work
- We test that every screen, report and endpoint that shows the figure returns the same value for the same period.
- Accuracy claims about classification, such as vehicle type, are measured on examples the model has not seen.
- Changes to a metric come with tests and a before-and-after comparison.
Frequently asked questions
Why does my average differ from a stopwatch?
A stopwatch measures one vehicle. AutoVision averages all completed sessions in the period, including quick stops and long stays, and shows the sample size.
Can the rules be changed?
The minimum and maximum session lengths are configuration settings. Any change applies to every screen at once, and the active rules are published with the figure.
Keep reading
- Pump occupancyPump occupancy analytics for fuel stations: how long vehicles occupy each pump, which pump is slowest, and when queues form. Honest, defensible numbers.
- Fuel station analyticsFuel station analytics from your existing CCTV: live pump occupancy, queues, ANPR and shop conversion. Built in Nairobi by DataFly Ltd.
- ANPR KenyaNumber plate recognition (ANPR) for Kenyan fuel stations: unique vehicle counts, pump-visit and shop links, repeat vehicles and hotlist alerts.