Uploading and reconstruction

Getting the card in

Open Surveys, click New survey, and drag the whole folder in. JPEGs straight off the card — do not rename them, do not resize them, do not run them through an editor that strips metadata. Everything the platform needs to place a photograph on the ground is in its EXIF, and an editor will happily discard it.

Upload runs in the browser and reads each frame as it goes: position, altitude, camera angle, capture time. It tells you what it found before anything is processed.

What it will tell you straight away

Frames with no position. These cannot be placed and will not be used. Usually a GPS that had not locked on the first few frames of the flight.
Frames shot off nadir. More than ten degrees off vertical. Set aside, not rejected — see Planning a flight.
Gaps in coverage. The footprints are drawn on a map as they load. Holes in it are holes in the reconstruction, and it is much cheaper to notice now than after an hour of processing.
Capture times. Shown in your own timezone. If a flight you know you did at four in the afternoon reads as four in the morning, the camera's clock is wrong — worth fixing, because it is what orders repeat surveys of the same ground.

What happens next

The survey queues, and a job runs:

Matching. Finds the same visible detail across overlapping frames.
Reconstruction. Turns those matches into a three-dimensional point cloud and works out where each camera was.
Surfaces. Builds two elevation models — a surface model, which is the top of whatever is there including trees and buildings, and a terrain model, which is the bare ground with those removed.
Orthomosaic. Reprojects every frame onto the terrain and stitches them into one flat, correctly-scaled map.
Tiles. Cuts the map into a pyramid so it draws quickly at every zoom.

You get an email when it finishes. You do not need to keep the page open, and closing the browser does not cancel anything.

How long

It depends on frame count, how much the frames overlap, and what else is running. Processing runs on a graphics card shared between every operator on the platform, up to four jobs at a time. A few hundred frames is usually well under an hour. A few thousand can be several.

The progress line tells you which stage it is in. If it sits on one stage for a very long time, that is worth reporting; it is not normal.

When it fails

The commonest failures, in order:

"Photogrammetry needs a set of overlapping images." The frames do not overlap enough to be matched. Fly again with more overlap, or over ground with more visible detail in it.

Not enough frames. A handful of photographs is not a survey. Reconstruction needs each piece of ground to appear in several frames from different positions.

A reconstruction that succeeds but looks wrong — warped, doubled, or with a piece of ground in the wrong place. Almost always repetitive ground: a uniform crop, a large body of water, fresh sand. The matcher finds the same-looking detail in the wrong place. More overlap helps; ground control points help more.

Everything is offset by a few metres. That is the aircraft's GPS, not the platform. Consumer drone GPS is accurate to a few metres absolutely, and much better relatively — so distances, areas and volumes within a survey are far more trustworthy than the absolute position of the whole thing. If absolute position matters, you need surveyed ground control.

Accuracy, honestly

Relative measurements within one survey — the volume of a heap, the area of a patch, the fall across a paddock — are good, typically within a few percent when the flight was properly overlapped.

Absolute position and absolute elevation are only as good as the aircraft's GPS, which is metres, and vertical is worse than horizontal.

None of it is survey-grade. If a number is going into a contract, a valuation or a regulatory submission, have a licensed surveyor confirm it.