Drone Data Accuracy


When capturing survey data with your drone, there are several things to consider to capture the best data possible on your site:

  • Image overlap
  • Flight speed
  • Ground control
  • Ground sample distance

Image overlap

Always make sure you’re flying with at least 70% overlap.

The basic principle of photogrammetry requires matching features between photos to form a single contiguous model. The images must overlap a lot for features to be matched. The biggest culprit we see that makes aerial photos unusable is the lack of overlap, and it’s often the easiest issue to fix.

In general, you need a minimum of 70% overlap between photos to ensure proper alignment for processing. This means that a third of its features should be new for each subsequent photo taken in a survey. Your flight control app should let you select the amount of overlap in your flights. If you need more help, we have articles on planning missions with the Phantom 4 RTK, Mavic 3 Enterprise, and the Matrice 300 RTK.

More tips

  • Make sure you’ve selected the right camera model.
  • Try to use a high point on your site for take-off. If you take off from a low point, you’ll fly low to the ground over higher areas and get poor overlap.

 

Flight speed

The accuracy of your data can be significantly affected by how fast you’re flying. Generally, you’ll need to fly slower for better accuracy and use your camera's highest possible shutter speed and ISO settings.

A typical fixed-wing drone like the Wingtra One flies at a minimum speed of 16 meters per second, and typically, a fast shutter speed is 1/1000th of a second. This means that by the time the shutter is open, your drone will have moved 16/1000ths of a meter — which is 1.6cm.

This movement can cause a very subtle warp/distortion in the whole image, an effect called "rolling shutter," resulting in reduced accuracy. Processing tools can try to account for this error. This movement can also cause blur in the image, heavily affecting accuracy.

A good rule of thumb is to avoid blurry images and a rolling shutter. You should use a shutter speed that means your drone has moved less than one-third of your GSD. If your camera can only capture at a shutter speed of 1/1000th and your drone can’t fly slower than 11m/s, you should plan your flight to capture just over 3.3cm GSD.

Ground control

There is no looking past ground control in the pursuit of accurate data. Other technologies, such as onboard RTK/PPK, will provide improved accuracy in the air, but this does not automatically translate to the same accuracy on the ground. Terrain models corrected with onboard RTK and PPK data are still more susceptible to turbulence, vibration, and camera calibration errors than models generated using ground control points.

Combined errors of ten degrees or more in a drone’s measured pitch or roll can introduce tens of centimeters of error in the final model if no GCPs are in place. With this in mind, achieving centimeter level ground accuracy in surveys with onboard PPK/RTK is significantly more difficult without the inclusion of ground control points.

Elevation of the site was surveyed seven times with AeroPoints compared to one flight with geotags only.

Absolute accuracy with GCPs can be refined to centimeters (not meters), making them the gold standard for survey-grade data.

Example planar shift between data sets generated with AeroPoints and those using just onboard geotags

Ground sample distance (GSD)

The GSD is the real-world size of a pixel in your images, and for most drone flights, it’s between one and 10 cm. A low GSD (1cm) is very high-resolution imagery, and a high GSD means lower resolution.

The GSD is determined by:

  • The resolution of your camera sensor — e.g., a 12MP camera will have a higher GSD than a 22MP camera
  • The altitude of your flight — flying higher means a higher GSD
  • The focal length/field of view of your camera — a wider angle camera like a GoPro means a higher GSD

Generally, a model's final accuracy in X and Y is one to two times the GSD and one to three times in the vertical. However, you will find that the average error tends to be closer to one times the GSD for all axes.

Accuracy in, accuracy out

Almost every aspect of a drone survey job — from the drone and camera used to the flight conditions and ground control implemented can affect the accuracy of the final data.

When submitting your data for processing, it’s important to understand that a 3D model is no more or less accurate than the data used to generate it. Consider accuracy requirements when setting up capture tools for the next flight, and you will be producing high-quality sources for your maps and models.


I still can't do it!

We wrote these articles to equip you with everything you need to get the job done on your own, but we understand that sometimes this isn't sufficient.

If you're stuck, you can connect with our support team by clicking the support button on the top right corner of your user portal.

Related to