A 2–5% error, and when that is enough
Automated bulk volume measurement does not replace scales. It gives continuity, and continuity is usually worth more than the precision of a one-off count.
The system builds a 3D surface model of bulk material from a video stream and calculates its volume. The error is two to five percent.
We state the number up front, because it decides whether the method fits.
- 2–5%
- volume calculation error
- continuous
- data update rate
- 0
- people in the measurement chain
Where it is enough
Tracking consumption. If the task is to see how much crushed stone left during a shift and whether that matches shipments, a five-percent error in both measurements partly cancels out, and the trend is clearly visible.
Monitoring stock. The question “will it last until the next delivery” does not need accuracy to the tonne.
Detecting discrepancies. This is where the method is strongest. A continuous series shows the moment stock changed without a document. A one-off stocktake cannot see that by definition.
Where it is not enough
Commercial measurement at dispatch. That needs scales, and no computer vision will replace them.
Settlements with a counterparty. A five-percent error on a batch is a dispute you will lose.
What improves accuracy
Known geometry. If the hopper dimensions come from laser scanning rather than being reconstructed from the same video, the error moves to the lower end of the range.
Stable lighting. The camera measures from the image, and sharp changes in lighting are a source of error. In a closed warehouse this can be solved. Outdoors it cannot.
Integration with weighing. Cameras over the loading point record how a dump truck body or railcar is filled. Together with the scales, that gives you material density, a value you cannot get any other way.
Which tool goes where
Closed warehouses: ceiling cameras watch hoppers and piles of sand, crushed stone, grain, ore. The data updates continuously.
Open stockyards: a drone on a schedule. Photogrammetry builds an elevation map, followed by automatic calculation of volumes and trends.
Conveyors: cameras along the belt measure the material profile: throughput, flow uniformity, and an alert when it deviates.
Loading: cameras over the point where truck bodies are filled.
The main argument
Not accuracy, but the absence of a person in the chain.
A manual stocktake is more accurate at the moment it is done, and it happens once a month. Between stocktakes you know nothing, and that gap is exactly where everything you care about happens.