How accurate are drone data? We compared it with a licensed surveyor’s data!

written by Dong Won Lee

Our DJI M300 RTK drone with a LiDAR sensor attached

This is a common question we get from our clients, and it’s a very good question for any client to ask before starting any work with a drone service provider, so expectations on deliverables are crystal clear. It’s important for the client to know how accurate the data will be. It’s also important for the service provider to know how accurate the data needs to be and what type of accuracy to achieve (absolute or relative).

When we were first starting out as a company offering drone mapping services, this was a tough question to answer. We were only able to reference other stateside companies’ work and the accuracies they’ve achieved with their equipment and workflows; so it was a tough sell. “Just because they've achieved this result, how can you be sure you can do the same?” would be a common worry… and understandably so!

Luckily for us, we had engineering companies who were early adopters and believers of drone technology, and our company went on to do many, many successful projects with various sensors for various clients in just a span of 18 months. Now, we are able to show you the accuracies we can achieve with drone data from our own equipment, our own “tested & tried” workflows and our own datasets.

Setting up our base station to collect raw GPS data

In this blog, we will show you the accuracies we’ve achieved in a photogrammetric model we processed; I think you’ll be blown away at the results. Before starting a huge project with an engineering firm, they wanted us to prove to them that our drone data can be accurate to a tenth of a foot (3cm) and we were up to the challenge! ----

Test Parameters

Method: Map (via photogrammetry) 1 mile of road and generate an accurate, colorized 3D point cloud

Goal: Achieve tenth of a foot accuracy (3 cm) against set features

Feature of interest: elevation (Z) data on manholes

Accuracy type: absolute accuracy

Reference system: Guam Geodetic Network ----

To briefly explain the difference between relative accuracy and absolute accuracy:

Relative accuracy is measuring a feature in relation to another feature. So for example, if you take a 20-foot container, it doesn’t matter whether that container is in Guam or in Saipan. We will still come out to a measurement of 20 feet. Or even the distance between two buildings. These are both relative measurements.


Absolute accuracy on the other hand is measuring the difference between a feature to a known point on earth; against some kind of reference frame. In Guam, the standard reference frame for surveying and engineering is called the Guam Geodetic Network.

One cool thing to point out is that models that are absolutely accurate are also relatively accurate. But models that are relatively accurate aren’t absolutely accurate.

Back to the accuracy test.

3D colorized point cloud | 1-mi road

The method for testing was simple. Map a 1-mi stretch of road. The surveyors would take a total of 9 ground control shots. They would give us 4 of those shots to use as our ground control points so we can georeference our drone data to those 4 points so we can achieve absolute accuracy. The remaining 5 ground shots were the manholes (feature of interest) and would serve as the checkpoints to measure the deviation between drone data and the surveyor's ground data. With this deviation, we can calculate the accuracy of our work. The goal was to have a deviation (error) of 3cm or better and we were very confident we can achieve this level of accuracy.



The results

After completing the processing of our photogrammetric model, we quickly looked to see how accurate our point cloud data was against the surveyor's 5 ground shots. Here are the results, point by point.

Point ID

Ground

Drone

Deviation (m)

Deviation (cm)

Check #1

98.284

98.297

0.013

1.3

Check #2

96.332

96.321

0.011

1.1

Check #3

95.302

95.295

0.007

0.7

Check #4

97.976

98.006

0.03

3

Check #5

99.332

99.315

.017

1.7

Point by point view from the drone point cloud data:



Conclusion

The results speak for itself. We can clearly see that the accuracies, when compared with the actual ground shots from a licensed surveyor, averages 1.56cm. It ranges as great as 0.7cm to 3cm. This just blows my mind how we can achieve centimeter-grade accuracy using drones today!


For any surveyors who may have doubts if this is even real, I challenge you to go out to this site and see the elevations for yourself. The coordinates are in the above photos! It was converted into WGS84 UTM Zone 55N so feel free to compare and see firsthand.

----

For any surveyors or engineering firms who would like to work with us, please contact us at dwl@2cofly.com | www.2cofly.com


For any aspiring drone pilots who wishes to start a new and exciting career, we have a full online course to teach you everything you need to know to pass the Part 107 knowledge test:

learn.2cofly.com


Other media




 






Dong Won Lee

Drone Ops | Co-founder

dwl@2cofly.com

(671) 988-1532

30 views0 comments