Advice on Ground Control Point Method for Site Calibration (Landscape Design Project)

Hi everyone,

I’m currently using RealityCapture to generate photogrammetry models as part of my work in landscape design, and I would appreciate some feedback on my current method for calibrating models using ground control points (GCPs).

Here’s how I’m proceeding:

  • I fly a drone to capture aerial imagery of the site.
  • On the ground, I’ve placed visible GCP markers prior to the flight.
  • For elevation data, I use an optical level (dumpy level) on-site to measure the height differences between the GCPs.
  • I also use a tape measure to determine ground distances between the GCPs to better understand their relative positioning.

Since I do not have accurate absolute coordinates (X, Y, or GPS values) for my GCPs, here’s the workaround I’ve been using:

  • After aligning my model, I identify the GCPs as tie points.
  • I then copy and paste their coordinates (as estimated by RealityCapture) into the Control Point setup.
  • Finally, I modify only the Z-value (altitude) of each point based on my measured heights on site.

My questions are:

  1. Is this a valid method for improving vertical accuracy when absolute georeferencing isn’t available?
  2. Could this introduce distortion or misalignment in the model that I might not be aware of?
  3. If this approach is problematic, what would be a better method for creating a reasonably accurate site model without survey-grade equipment?

My end goal is to generate a site model accurate enough for conceptual design and terrain analysis in residential projects.

Thanks in advance for your time and insights.
Frederic
Landscape Architect – NSW, Australia

Hello Frederic,
I can imagine using such workflow for not so precise works.
There can be some issues mostly with the positions, as the generated model can have some distortion. It all depends on the used drone, the way of capturing and other factors.
The best option here is to measure such control points using the valid surveying method.