Solutions — 08
AI Counting & Volume Measurement
How much is there? How many are there? Two questions that still send somebody out with a folding rule and a clipboard.
A survey flight produces a terrain model. From it the DroniumApp calculates volumes: gravel stockpiles, bulk material, excavation, landfill bodies — stated with their tolerance, not as a round number. For unit counts, a detection model counts what stands on the area: vehicles, containers, pallets in open storage, trees in a stand.
The difference from a manual survey is not primarily accuracy but repeatability. Two people measure the same stockpile differently; the same flight analysed twice yields the same result. Only then can two points in time be compared at all.
Why manual surveys stay imprecise, however carefully they are done
Measuring a bulk stockpile by hand means pacing the base area, estimating or spot-measuring the height, assuming a slope angle and calculating the volume. Every one of those steps carries an error, and the errors add up. Between two surveys of the same pile there are typically several per cent — with six-figure material values that stops being a rounding error.
Then there is the effort. A large area costs half a day, and because it costs that, it is surveyed twice a year rather than monthly. Between surveys nobody knows exactly what is there.
Counting has the same problem in another form. Vehicles on a yard, containers on a transfer area, logs at a landing: countable, but not twice the same, and nobody enjoys doing it twice.
How the measurement works
The basis is a terrain model produced photogrammetrically or by LiDAR. The area is flown in a fixed grid, the overlaps produce a point cloud and from it a surface model. The volume follows as the difference between that surface and a defined reference plane — the formation level, the previous state, or a surveyed base.
Which reference applies is the real technical question, and it is settled once, up front. Without it any software returns a figure nobody can check.
For unit counts a detection model is used, trained on your imagery. A model that counts containers has to know what your containers look like, in which sizes they occur, and when two standing side by side count as two rather than one.
What the figure is worth
Every measurement is reported with its tolerance. A volume figure without a tolerance is an assertion — and in a dispute with a supplier, the one that does not hold.
The tolerance depends on flight overlap, ground resolution, control points and the nature of the material. Fine sand can be captured more accurately than coarse ballast with voids. What is achievable, we say before the flight, not after.
Where it pays off
Wherever quantity is money and the quantity is currently estimated. Gravel works and quarries for stock valuation, contractors for billing excavation and fill, landfills for remaining-capacity planning, logistics operators for yard utilisation, forestry for stand inventories.
The second gain is frequency. What costs half a day is done rarely. What is flown in an hour and analysed automatically can run monthly — and then you see not just the quantity but its development.
→
Benefits at a glance
- Repeatable results: the same flight yields the same figure twice
- Every measurement with a stated tolerance instead of a round estimate
- Unit counts from a model trained on your own imagery
- Measurable more often, because the analysis is not done by hand
→