The Technology Behind Precision

We fly an RTK-equipped DJI Mavic 3 Enterprise. RTK — Real-Time Kinematic GPS — uses a fixed base station that broadcasts position corrections to the drone in flight, pinning every photo to a position accurate to roughly 1 centimeter horizontally and 1.5 centimeters vertically. That's the foundation of the 2-3% volumetric accuracy we deliver on stockpiles, versus the 5-15% you'd get from a non-RTK consumer drone.

Every flight is conducted by an FAA Part 107 certified pilot under $2M of liability coverage. Flight planning is computed per site: photo overlap (typically 75-80%), altitude (low enough for resolution, high enough for coverage), ground speed, and the location of the RTK base station relative to the work area.

Note: stockpile volume is a relative measurement. We measure the base of the pile and the top of the pile in the same flight, with the same RTK lock, so any systematic position error cancels out. The relative precision RTK provides is exactly what the volume math needs.

Once the photos land, our processing pipeline takes over. We provision a dedicated Hetzner server for each job — your data never shares CPU or memory with another customer's flight. The server runs NodeODM, an open-source structure-from-motion engine that turns 200+ overlapping aerial photos into a 3D model of your site.

The pipeline runs in five stages: feature matching (identifying the same point on the ground across multiple photos), bundle adjustment (solving for the position of every photo simultaneously), dense reconstruction (computing depth at every pixel), mesh generation (stitching depth into a continuous surface), and texture projection (painting the photos back onto the surface). The result is a textured 3D model tied to real-world coordinates.

From the 3D model we generate a normalized Digital Surface Model — an nDSM — which is essentially a heat map of how high every point on your yard is above the surrounding ground. Pile heights are isolated from ground level by filtering out the lowest points around each stockpile footprint, fitting a base surface, and measuring everything above it.

Volume is then computed by integrating the height grid over the pile footprint: for each grid cell inside the boundary, we know the area and the height above the base, and the product of the two summed across the whole pile gives us the cubic yards. The math is the same math a surveyor would do with a tape measure and a calculator — just on millions of points instead of three.

Cubic yards become tons via material-specific density. The platform stores per-material density values for every material on your site — concrete sand, washed gravel, #57 stone, RAP, and so on. Defaults are sourced from typical aggregate density references, but every density is editable per company, because your quarry's #57 stone may not weigh the same as the next quarry's #57 stone.

When you hand us scale tickets, we can also calibrate density empirically: if your loaded weights consistently come in higher or lower than the density assumption, we surface that drift back to you so you can update the density value for future flights.

The flight tells us what's physically on the ground. Your scale tickets tell us what you've sold and what's been delivered. Reconciliation is the comparison between the two — and the gap is where your money lives.

You drag and drop your scale ticket exports into the dashboard (or we pull them from your scale software directly). The platform reconciles them against the measured inventory and surfaces drift in either direction: over-scaled means inventory shortfall (you've sold more than you actually had), under-scaled means phantom inventory (your books say more than the drone can find). Either way, you see the gap in dollars, by material, in the month it shows up — not at year-end.

Here's the most important thing to understand about what we sell: most aggregate yards never physically measure their inventory. They trust scale tickets, watch the piles, and reconcile at year-end. The 3-5% drift that quietly accumulates every quarter is the result.

We give you a physical, independent measurement of every pile in your yard — accurate to 2-3%. That's not perfect. But it doesn't need to be. The drift is bigger than the measurement error, which is exactly why the drift is detectable. Verification beats estimation, even imperfect verification.