Stockpile Volume Measurement AI

Add an analysis layer to your stockpile surveys with vision AI for stockpile volume measurement. Built for the sites where the flight takes twenty minutes and the boundary tracing takes the rest of the day, where two surveyors draw the same toe line differently and the inventory number moves with them, and where the material sitting under a shed roof gets estimated because nothing can fly in there. Whether you're managing run-of-mine pads, quarry and aggregate yards, port stockyards, or cement and coal storage, Roboflow works on the imagery your site already captures and takes the repetitive judgment out of the survey.

Segmentation, Toes, and Classification:

• Segment each pile from the pad and from its neighbours, and delineate the toe consistently rather than differently for every operator and every survey
• Classify material type per pile, so ore grades, aggregate products, and coal qualities match to the right inventory record automatically
• Count piles and map the yard, so what is on site is known between surveys rather than at them

Change, Trend, and Covered Storage:

• Detect what changed between surveys: piles built, drawn down, moved, merged, or newly created since the last flight
• Monitor level and change in sheds, domes, bunkers, and bays from fixed cameras, where a drone cannot fly and the alternative is an estimate
• Trend pile activity continuously between formal surveys, so a drawdown is visible in days rather than at the next month end

Records, Reconciliation, and Audit:

• Keep a dated, image-backed record of every pile with its segmentation and classification, so an inventory figure has evidence behind it
• Feed consistent boundaries and material calls into your existing photogrammetry, survey, and ERP workflow rather than replacing it
• Support month-end reconciliation and audit with a defensible trail of what was on the pad, when, and how it was measured

Bring intelligence to every survey today. Stop a hand-drawn toe line from becoming an inventory variance, a covered shed from becoming a guess, or a month-end close from waiting on a queue.

‍

More About Stockpile Volume Measurement

Where does the toe line problem cost money?

In the variance. A stockpile's volume is calculated between its surface and a base plane, and the toe is where the pile meets the pad. On a well-defined conical pile that boundary is obvious; on a long ROM pad with material feathering out, piles merging into one another, spillage, and a pad surface that has itself moved since the last survey, it is a judgment call. Two surveyors will draw it differently, and the difference propagates straight into the volume and then into tonnage after density is applied. Because that judgment is repeated every survey by whoever is available, the number can move when the operator changes rather than when the pile does. A segmentation model applies the same boundary logic every time, which does not make the toe objectively correct but does make it consistent, and consistency is what a month-over-month reconciliation needs.

What about material under a shed or in a bunker?

This is the gap in most stockpile programs, and it is where fixed cameras earn their place. Drones do not fly inside a storage dome, a coal shed, or a covered aggregate bay, so that material typically gets estimated, and it is often a significant fraction of what is on site. A fixed camera watching a bay cannot produce an absolute volume any more than a drone photo can, but it can track relative level and change continuously, detect when a bay is filled or drawn, and give a trend between the physical surveys that do happen. Paired with a periodic laser scan or survey to anchor the absolute figure, that turns a monthly guess into a monitored number. Being precise about the split matters: continuous relative change from vision, absolute measurement from the survey.

Can it integrate with our survey, GIS, and ERP systems?

Yes. Segmentation masks, toe polygons, material classifications, and change detections export in standard geospatial formats and push into survey and photogrammetry pipelines, GIS, and inventory and ERP systems through REST and direct database writes, keyed to your pile and stockpile identifiers. Results carry the survey date and the imagery behind them, so a figure in the inventory system traces back to a specific flight and a specific frame. Processing can run in your VPC or on-prem, which is the normal requirement when survey imagery covers an operating site, and inference can run at the edge where remote pads have limited connectivity.

‍