Turning aerial imagery into maps, models and measurements people can rely on

What this actually solves

Ground survey is expensive, slow and, on a live site, genuinely dangerous. A large parcel gets walked and recorded point by point. Stockpiles get measured by walking around them. Progress on a construction site gets recorded by someone with a camera and an opinion. All of it is work that has to be repeated every month, and because that cost lands again every cycle, how often you measure ends up set by the budget rather than by what the decisions need.

Aerial capture collapses the field time and produces something denser than a crew could ever record by hand — but the imagery itself is not the product. What you need is a number. Mining companies get paid against extracted volume. Contractors bill against progress. Land disputes turn on area. Agricultural decisions turn on which part of a field is failing. The photographs only become worth paying for once they have been turned into a map that is correctly positioned, a model that is measurable, and a figure someone is willing to put their name to.

That last part is the whole risk in this service, and it is worth stating up front. A beautiful orthomosaic and a worthless orthomosaic look exactly the same on screen. The difference is whether the imagery was tied to the ground properly during capture — surveyed ground control points on the site, or an aircraft flying RTK against a base station. Without one of those, the aircraft's own positioning is the only reference, and it is not tight enough to measure against. A map built on it renders perfectly, zooms beautifully, and gives you an area that would lose an argument in front of anybody who checks.

How the work runs

1. Establish what is being measured and to what accuracy. Before any file moves. Are you buying a visual record of progress, or a volume that money changes hands over? That answer sets whether ground control is mandatory, what the coordinate system and projection must be, and what the accuracy statement on the final report will say. If the imagery already exists, this is also where we ask in writing whether GCPs were placed and whether RTK was used. Half a day to two days.
2. Plan the capture, or audit the imagery that arrives. For a fresh capture that means altitude, flight pattern and overlap — enough overlap that every point on the ground appears in many frames, which we plan at around seventy-five percent front and sixty-five percent side and raise over surfaces with little texture — plus permissions and the control layout on the ground. For imagery that has already been shot, the same properties are checked from the images and logs, and any gap is reported before processing time is spent on it. Two to five days.
3. Photogrammetry processing. Alignment, dense reconstruction, then the outputs — orthomosaic, point cloud, digital elevation model and mesh — in Agisoft Metashape or Pix4D, with WebODM where an open toolchain is preferred. This stage is mostly machine time and it is not linear in image count, so it is scheduled rather than squeezed. The reconstruction is inspected for holes and warping before anything downstream is built on it. Three to seven days.
4. Measurement and analysis. QGIS for map work, areas and elevation products; CloudCompare for point cloud comparison and volume between two surfaces. Volumes are computed against a stated base surface, and the choice of that base surface is written down, because two honest people using different bases will get different numbers from the same cloud. Two to five days.
5. Web delivery. The heavy rasters are tiled — gdal2tiles or an equivalent pipeline — so the browser only fetches what is on screen, and the point cloud is served through Potree or Cesium. The viewer carries distance and area measurement, and a two-date comparison where the site is being tracked over time. Hosted on object storage so it stays cheap to keep online after handover. Four to eight days.

What moves the price

The range is $1,000–$15,000 across 2–8 weeks, and it separates on three things.

Whether flying is part of the job. Capture brings scheduling, weather, site access, permissions and ground control placement, and any one of those can turn a booked day into a wasted one. When usable imagery arrives already captured, the work starts at processing and the range sits much lower.

How strict the measurement has to be. A monthly visual progress record and a volume figure that a payment certificate depends on are different products from identical-looking files. Survey-grade output means control, verification against known points, and a written accuracy statement — more time, and more responsibility carried.

Site size and surface. Image count drives processing hours, but texture drives risk. Bare earth and rubble reconstruct well. Fresh concrete, standing water, glass roofs and uniform crop reconstruct badly at normal overlap, and either need denser capture or produce gaps that have to be handled honestly in the deliverable.

What usually goes wrong

Measurement liability is accepted without knowing how the imagery was captured. Someone sends a folder of images, a good-looking orthomosaic comes out, an area is reported, and it turns out no ground control was placed and the aircraft was not RTK. The map is out by more than the measurement can carry, and you have already invoiced against it. The protection is a written question at the start of every project — GCPs, RTK, expected accuracy — and, when the answer does not come, stamping the deliverable visual reference only, not survey grade rather than letting anyone assume otherwise.

Overlap is set too low for the surface being flown. Photogrammetry works by finding the same features across many overlapping frames. Cut the overlap to save flight time, or fly a surface with nothing distinctive on it, and the software cannot tie the frames together — the point cloud comes out with holes exactly where the featureless ground was. It is only discovered after hours of processing, and there is no software fix. It is a reflight, which is why overlap is a planning decision and not something to economise on.

The raw files are handed over as the deliverable. The GeoTIFF and the point cloud come out far too large to open on an ordinary machine. Either you never look at what you paid for, or you load it into whatever software you have, which downsamples to cope, and then you measure on the downsampled version and get a number that disagrees with the report. Tiled delivery in a browser viewer is not a presentation nicety here — it is what keeps everyone measuring the same thing.