What is the relative and absolute accuracy of drone mapping

The article provides definitions of the relative and absolute accuracy of a photogrammetric project. It also points out how using Ground Control Points or RTK/PPK drones increases the absolute accuracy of results. 

The accuracy of the outputs can be distinguished into:

  • Relative accuracy of a photogrammetric project is defined by comparing individual features on a map / reconstructed model / orthomosaic with other features on the same model. For example, the relative accuracy of the model is high if two points have the same distance between them on the model, and in reality. 
  • Absolute accuracy of a photogrammetric project is defined by the difference between the location of features on a map / reconstructed model / orthomosaic and their true position in a certain reference frame. For example, if the position of a road in the reconstructed model is close to its actual position on the Earth, then the absolute accuracy is high. Adding precisely measured GCPs or using RTK/PPK drones can greatly improve absolute accuracy.

High relative accuracy, low absolute accuracy

The relative accuracy of the reconstructed model is high as the dimensions are correct. For example, the height of the building is 5 meters on the model and in reality.
The absolute accuracy of the model is low as it is not georeferenced correctly. For example, when loading the model on a precisely georeferenced map, the house is located in the middle of the road.

 
Information:
   

Relative accuracy

The relative accuracy will depend on the quality of the reconstruction of the project, which itself depends on the overlap between images, the visual content of the images, and many other parameters.

Generally, one can expect an error of 1-3 times the Ground Sampling Distance (GSD) of the original images for the relative position of a point in a project that is correctly scaled and reconstructed.

 
Important: This error is not global. Some points will be more accurate than others based on the number of images on which they are visible (overlap), how close they are to manually introduced points, the GSD in this area, etc.
 
Note: This error can be reached on any scale. However, for sub-centimeter GSD projects, it is harder to get the ideal conditions. Problems such as perspective, vibrations of the camera, blur, depth of field, etc. are magnified. It is recommended to test and plan for such projects, in order to get the expected accuracy. Rolling shutter cameras are not recommended for sub-centimeter accuracy results.

To increase and assess the relative accuracy we recommend using Manual Tie Points (MTPs), scale constraints, and orientation constraints.

Absolute accuracy

Absolute accuracy highly depends on the accuracy of the GNSS receiver of the drone and if GCPs are used in the project or not. 

For example, drones with a standard GNSS receiver provide image geolocation with an expected accuracy of a few meters. If no GCPs are used in the project to precisely georeference it, then the final absolute accuracy of the reconstructed model is also expected to be in a range of a few meters

To reach cm-grade accuracy, we recommend using:

  • Drones that provide RTK/PPK image geolocation.
  • Ground control points.
 
Important: The absolute accuracy of the results cannot be higher than the accuracy of image geolocation, or GCPs' accuracy.

Generally, the expected absolute accuracy of a correctly reconstructed and georeferenced project is:

  • 1-2 x GSD horizontally (X,Y coordinates). 
  • 1-3 x GSD vertically (Z coordinate).
 
Example:

For a project flown with a drone using a standard GNSS receiver (accuracy range of a few meters), with a GSD of 5 cm, the expected:

  • Relative horizontal accuracy is 5-10 cm and the vertical accuracy is 5-15 cm.
  • Absolute horizontal and vertical accuracy is in the range of the GNSS receiver accuracy (few meters of accuracy)

For a project flown with a drone with a standard GNSS receiver, with a GSD of 5 cm, and GCPs the expected:

  • Relative horizontal accuracy is 5-10 cm and the vertical accuracy is 5-15 cm.
  • Absolute horizontal and vertical accuracy is in the range of the GCPs errors.

For a project flown with a RTK/PPK drone, with a GSD of 5 cm the expected:

  • Relative horizontal accuracy is 5-10 cm and the vertical accuracy is 5-15 cm.
  • Absolute horizontal and vertical accuracy is in the range of the RTK/PPK drone accuracy.
Adding ground control points (GCPs) improves both the relative and absolute accuracy of the project. In addition to MTPs, GCPs include the measured coordinates which are used for georeferencing the project.

We recommend using checkpoints to assess the absolute accuracy of the model. With checkpoints, it is possible to get the difference between the reconstructed model and the known position of points For more information see Difference between a ground control point and a checkpoint article.

What is the Point cloud, DSM, and orthomosaic accuracy

The point cloud, DSM and, consequently, the orthomosaic accuracy are also affected by the quality of the initial images and their visual content. Sharp edges, trees, reflective surfaces, and certain types of roads and rooftops may be locally less accurate.
Was this article helpful?
18 out of 22 found this helpful

Article feedback (for troubleshooting, post here)

3 comments

  • Takeyuki Suzuki

    In my sub-centimeter GSD project, polyline's length doesn't show sub-centimeter value. How can I show it?

  • Ina (Pix4D)

    Hi,

    At the moment in the mapper, you can measure distances at the minimum of cm level.

    Having measurements under centimeters would require accuracy at that level as well. Having consistent subcentimeter accuracy it might be challenging. The accuracy of your outputs you should expect is 1-2 pixels (GSD) horizontally (X, Y coordinates) and 2-3 pixels (GSD) vertically (Z coordinate). The GSD of the data set depends on the focal length, the "flight" height, the sensor and the image height. So, if you want to acquire a sub-centimeter accuracy, then your GSD should be half that. So, you should design the "flight" plan in such a way to acquire the adequate data for obtaining such an accuracy.

    Thanks,

    Cristina

  • Takeyuki Suzuki

    OK. Thank you, Cristina.

Please sign in to leave a comment.