Summary: Using the georeferencing tool included in Site Scan Manager, it is possible to generate orthomosaics and DEMs accurate to their theoretical limit, two times GSD or 1.5" in this case, by setting only 5 GCPs across a site in a scattered pattern.
Accurate orthomosaics, DEMs, and contours are important for progress monitoring, civil estimations, and pre-construction design. However, the GPS receivers and gimbal controllers found on small UAS only guarantee accuracy to around ten feet. To bring your orthomosaics into your project coordinate system and use the data to compare as-planned and as-built, inch-level accuracy is required. To get there, Ground Control Points (GCPs) are required.
Prior to the 11/30/16 Site Scan Manager launch, applying GCPs to orthomosaics required complex desktop GIS software and the associated expertise, but now georeferencing is a few clicks away in your web browser. Complete instructions to upload GCPs and georeference orthomosaics are available in the Site Scan Knowledge Base.
However, it is important to understand what level of accuracy is available after georeferencing using the easy Site Scan Manager tool. To determine this, the the 3DR team worked with a customer to perform a survey of a potential build.
Using an EOS GNSS Arrow 200 RTK rover and the CRTN network, we laid out 10 GCPs across the approximately four acre site. Before each point was captured, we ensured the RTK solution had converged to a centimeter-level accuracy.
We then flew the site twice at 200 feet, giving us a Ground Sampling Distance (GSD) of a less than an inch. This can be verified by magnifying the center of the target and counting the number of pixels spanning a known distance. The black stripe in the middle of the target measures 1 7/8" and spans almost three pixels, yielding a GSD of 0.7".
As expected, the location of the targets in the raw orthomosaic deviated from their actual positions by several feet. However, we wanted to learn what kind of accuracy we could achieve with the Site Scan Manager georeferencing tool.
First, we used nine GCPs and held either Point 10, off in the corner, or Point 9, smack in the middle, as checkpoints. We then measured how far the surveyed GCP was from the center of the target, as shown in the image below. This yielded four data points across the two jobs for a mean error of 1.4 inches +/- a standard deviation of 0.8 inches.
Next, we wanted to learn whether we could reduce the number of GCPs while maintaining the same level of accuracy. We processed both flights with either Points 1, 3, and 7 or Points 1, 3, 5, 7, and 8 as checkpoints, leaving 7 or 5 GCPs, respectively. With 7 GCPs, we determined the mean offset plus or minus the standard deviation was 1.4" +/- 0.5" and with 5 GCPs 1.5" +/- 0.6". These numbers are all well within their errors, indicating that a site this size requires no more than five GCPs to reach theoretical limit of accuracy.
Raw data are found in the table below. If you would like to get hold of the sample images and GCPs, please contact us and one of our Solutions Engineers will walk you through the methodology.
|GSD = 0.7 inch||9 GCPs (offset in inches)||7 GCPs (offset in inches)||5 GCPs (offset in inches)|
|Job 2, Point 10||1.8||Job 1, Point 1||1.6||Job 1, Point 1||1.5|
|Job 1, Point 10||2.5||Job 1, Point 3||0.9||Job 1, Point 3||1.1|
|Job 2, Point 9||1||Job 1, Point 7||0.9||Job 1, Point 5||0.7|
|Job 1, Point 9||0.3||Job 2, Point 1||1.9||Job 1, Point 7||1.5|
|AVERAGE||1.4||Job 2, Point 3||2||Job 1, Point 8||1.5|
|STDEV||0.8||Job 2, Point 7||0.8||Job 2, Point 1||2.4|
|AVERAGE||1.4||Job 2, Point 3||1.8|
|STDEV||0.5||Job 2, Point 5||2|
|Job 2, Point 7||0.5|
|Job 2, Point 8||2|