Through our photogrammetry and CAD linework service at Aerotas, we have the benefit of seeing thousands of drone projects. Combined with our own extensive testing, this has allowed us to learn the key practices for ensuring drone surveying is as accurate and efficient as possible.
One of the most critical factors in ensuring a profitable and accurate drone survey is ground control. When using a drone for survey work, that’s where accuracy begins. “Ground control points” (GCPs) are targets marked on the ground and located to a high degree of accuracy with traditional surveying tools such as high-accuracy GNSS, total stations or laser scanners. GCPs are then used to reference and correct the 3D model created from the drone photos, ensuring that the model is accurately georeferenced and matches field shots taken on the ground.
If drone photos are stitched together without GCPs, the resulting 3D model is like a rubber sheet that can easily stretch or warp by as much as 10-feet in the X and Y dimensions and 50-feet in the Z dimension. This is due to inaccuracies in the drone’s onboard GNSS and noise caused in the photo-stitching process.
Incorporating GCPs in the model is like driving nails through that rubber sheet, anchoring the final data to a known set of coordinates on the ground. This ensures that all data is in the same coordinate system and that the 3D model is highly accurate within that coordinate system. Even when using a drone with high-accuracy GNSS (like onboard RTK or PPK), GCPs are still required to correct for errors and ensure accuracy.
Though projects vary widely, we have found some simple benchmark guidelines that can help in the project planning process. Geometrically, three GCPs are required to correct a model, but we recommend a minimum of five, no matter how small the project. This allows the photogrammetrist to check GCPs against each other and make corrections for any errors that occur in processing.
Many projects will need more than five GCPs. However, it’s not a simple GCPs-per-acre calculation. What matters is how many photos “bridge” between GCPs. If the drone is flown low, each photo will cover a small amount of ground, so it takes a large number of photos to bridge between two GCPs. When flying higher, however, each photo covers more ground, so fewer photos bridge between the same GCPs. The more photos that bridge between those GCPs, the more flex and error there will be in the model. This means that when flying at a low altitude, more GCPs are needed than if the same site was flown at a high altitude.
Ideally, the number of GCPs should be based on the number of photos between the GCPs. In order to simplify this, we base the number of GCPs on batteries used. From our testing with the DJI Phantom 4 Pro, five GCPs per battery is an effective standard. On average, this equates to five GCPs for each 250 photos taken. For example, covering 50 acres when flying 400-feet above ground level (AGL) would only require five GCPs, because it can be flown in a single battery. However, covering the same 50 acres at 100-feet above ground would require 10 batteries, and thus, 50 GCPs.
The number of GCPs required drops significantly as flight altitude increases. In practice, it is impractical to fly large project sites at low flight altitudes. Covering 200 acres could be done in only four flights and 20 GCPs at an altitude of 400-feet, while at 100-feet altitude, it would take 10 times that – 40 flights and 200 GCPs. However, there is a tradeoff: flying higher will produce lower expected accuracy.
|Site size||Flight altitude||Expected number of batteries||Expected number of photos||Required GCPs|
Predicting, measuring, and validating accuracy from drone data is very complex and different on every project. The only way to independently and perfectly measure accuracy for each project is by using checkpoints – independent field-shots measured against the model produced from the drone data. However, through our testing, we have determined benchmarks for expected accuracy when following proper procedures.
Accuracy is impacted most by the altitude flown. Adding more GCPs beyond the recommended five per battery does not dramatically improve accuracy in our testing. The data in Figure 2 is based on testing with a DJI Phantom 4 Pro.
As can be seen, there is a clear tradeoff between accuracy and flight altitude. This means that careful project planning is essential to ensure a drone survey produces the needed accuracy and does so efficiently. Batteries used and GCPs both translate to more time spent in the field, so always flying for maximum accuracy is often not the best business decision. Often it makes more sense to rely on the drone for topo data (contours) and then collect any critical high-accuracy points on the ground.
|Flight altitude (ft. AGL)||Photo resolution (GSD)||Maximum area covered per 17 min flight||Expected horizontal accuracy (ft.)||Expected vertical accuracy (ft.)|
|100-feet||0.85 cm / 0.03’||5 acres||<0.1||0.10|
|200-feet||1.7 cm / 0.06’||20 acres||0.14||0.20|
|300-feet||2.55 cm / 0.08’||35 acres||0.21||0.29|
|400-feet||3.4 cm / 0.11’||50 acres||0.28||0.39|
At a minimum, GCPs need to be set in each corner of the project area, and at least one in the middle. If there is a significant amount of elevation change on the site, ensure that there are some GCPs set on the high and low points of the project. Beyond that, best practice is to have the GCPs spread evenly throughout the project.
Next, it is important that the flight area is set to extend beyond the GCPs by at least one flight-line or photo in each direction. If a GCP is outside of the flight area, then it cannot be included in the image-stitching process and will cause high error in that area of the project.
When conducting a linear project (corridor, pipeline, right-of-way, etc.), best practice is to set pairs of GCPs along either side of the corridor at an even interval. For example, if flying at 400-feet, a pair of targets about every 500 to 700 feet is sufficient to meet the above accuracy expectations.
Finally, GCPs need to be clearly marked, so that the exact point that was located in the field can be matched in the photos. This means that the GCPs need to have a clear, unambiguous point so the photogrammetrist can be confident that they are entering the coordinates at the exact same point that the surveyor shot in the field. In our experience, checkerboard-pattern targets work best for this purpose.
It’s also important that targets be easy to identify in as many photos as possible. This means that they must be:
- Away from any obstructions (i.e. trees or buildings) that would block a clear view from the sky from every angle.
- Large enough to be seen in the drone photos (at least 12 inches across).
- Easily identifiable from any orientation (i.e. not “left side of paint stripe”).
- A color or contrast that stands out.
Ground control is where accuracy starts with drone surveying, and it’s what ensures the model fits into the coordinate system you need. It is a crucial step in the project planning process, and requires thorough understanding of tradeoffs for accuracy and field-time. Ultimately, a drone survey being accurate and profitable depends on the right amount of the right type of GCPs set in the right locations around a project site.