Ground control: marking targets in the images

Put a reconstruction on your own survey instead of the drone's GPS, by pointing at each target in the photos.

7 min read

Why marking matters

A drone records roughly where each photo was taken from its own GNSS. That is enough to put a model on the map, and not enough to put it on your control. Surveyed ground control fixes that, but only if the processor is told where each target appears in the photographs, not just where it is on the ground.

That is what marking is: for each target, you point at it in the frames it shows up in. Backsight records the pixel, writes the control file the processor reads, and the reconstruction is placed on your survey.

The whole workflow, in order

Survey your targets in the field, on the project's coordinate system.
Create the mission and upload the flight's images (step 1).
In step 2 choose Use surveyed control, then add the points, type them, or import them from a point dataset, marking which are control and which are held back as checks.
Below the points, mark each control target in at least three frames; the step says when the targets are ready.
Request the reconstruction. The control file is written for you and sent with the job.
When the products come back, check them against the check points you held back.

Note

Marking happens after the images are uploaded and before the reconstruction is requested. Nothing is charged until the job actually starts, so you can take your time over it.

Marking a target

Step 2 of the mission, under the control points.
Choose the target from the list. Its surveyed coordinates are shown beside it, so you know which one you are looking for.
Step through the images with Previous and Next, or pick one from the list, frames you have already marked carry a tick.
When you find the target, click on it. A crosshair appears where you clicked.
Zoom in and nudge with the arrow keys until the crosshair is on the centre of the target. Shift and an arrow moves ten pixels at a time.
Save the mark. Backsight moves you to the next frame so you can keep going.

Accuracy is worth the extra few seconds

A mark two pixels off is a real error in the adjustment. Zoom to 4× or 8× before you save, and put the crosshair on the exact centre of the target, not near it.

How many, and how many frames

What	Minimum	Better
Control targets marked	3	5 or more, spread across the site
Frames per target	3	4 to 6, from different angles
Check points held back	0	2 or more, away from the control

Three targets is the minimum that can fix a model in space, and a target seen in fewer than three frames is not constrained. Backsight leaves an under-marked target out of the control file rather than letting it pull the adjustment, and tells you which ones still need frames.

Spread them out

Three targets in a line, or all in one corner, will place a model badly however carefully you marked them. Put control at the edges of the site and at least one in the middle.

Control against check

A control point is used to place the model. A check point is held back and used to measure it afterwards. Backsight never writes a check point into the control file, feeding it to the adjustment would destroy the only independent measure you have of the result.

To check a finished reconstruction:

Build a surface from the returned point cloud.
Surfaces → Check against points, and paste your check shots.
Read the residual at each one: surveyed minus surface.

What gets sent, and what the job records

A control file listing every marked observation: the surveyed coordinate, the pixel, and the image it is in.
The projection your control is on, taken from the mission.
A note on the job saying how it was georeferenced, and from how many points and observations.

If fewer than three targets are properly marked when you request the job, no control file is sent and the reconstruction is georeferenced from the images' own GPS, with the job saying so plainly rather than implying control was used.