The target of every aerial surveying and mapping project is to create models and drawings that fulfill the accuracy standards of surveyors and engineers. It can be very frustrating if accuracies in a UAV mapping projects don’t add up. Fortunately, errors can often be avoided by sticking to a few essential rules. Between the past 2 years of mapping project at AirWorks and the team’s 5 years of experience with UAV mapping, we’ve narrowed down the best techniques for producing accurate maps, as well as many common sources of error.
For example, through many trials and errors, we’ve found that the ground sampling distance needs to be at least 1in/ px and that we need to validate that the average error within the x, y, z of all available control points is under 0.1 ft in order to achieve the highest accuracy. We’ve also learned that most errors can be avoided by sticking to a few essential rules. Here we will outline the most common errors we’ve noticed and how to make sure that you avoid them on your project.
1. Not using any ground control points (GCPs)
Why It’s a Problem
The 3D model will be accurate in x and y to the precision of the drone GPS, which usually means an accuracy of around 5 meters. This means the relative error of dimensions is therefore small horizontally--probably within a few inches--but the absolute error to an object’s real true GPS location will be 5 meters or more. It gets worse for vertical information. We have seen models floating in space, vertically deviating by a couple hundred feet from their true location. This deviation is different per site, vertical data in absolute terms is therefore not usable without GCPs.
How to Avoid It
Use some sort of ground control – a survey control station such as the Leica GS16 or GS18, or control points specifically designed for the purpose of mapping such as aero points.
You can also contact AirWorks to get our signature targets!
2. The ground control layout is not optimized.
This could mean your ground control layout doesn’t include enough coverage on the edges of your site or perhaps is too sparse in areas that are segregated by trees.
Why It’s a Problem
If there are no ground control points on the edges of the property, the computer doesn’t have any information to geo-rectify a site in those regions. Also, there won’t be any check points to actually make sure that you’re accurate. It’s the worst scenario, not only can you not assure accuracy, there is virtually no way to measure how far you are actually off.
If there a section with tree coverage, often the photogrammetry software cannot find matches within the imagery due to shifting tree canopy, and the point cloud in those areas will be black. As a result, there will almost be 2 different models created, one before the tree line and one after the tree line. If each section doesn’t have at least 3-4 points, the entire section won’t be anywhere close to its real location. Such an error has caused us at least 1 re-fly.
How to Avoid It
Always use too much instead of too little control- you can always discard a point, its hard to add one afterwards.
Spread the points evenly across the site and use back-up points along the edges of the flight path to use if a point gets disqualified for whatever reason.
Make sure you account for the site geography when designing the control layout, especially tree lines, but also highs and lows.
3. GCP accuracy measurements are erroneous.
When the error measurements on your points are higher than 0.1ft, it likely indicates something went wrong with your ground control points. Some causes of erroneous readings that we know of include low satellite coverage, or the point has been physically moved between being measured and doing the flight.
Why It’s a Problem
The only thing worse than no ground control is bad ground control.
If erroneous ground control points are being used during the processing it will be impossible to achieve a project accuracy of 0.1 foot -- and worse, it’ll be very frustrating to try to find out which points the errors come from.
There is an infinite amount of combinations of ground control and check points, and for every combination you might be looking at hours of data processing. We have wasted days of processing during the photogrammetry step until we established means to avoid these errors. There have even been times when we've had to re-fly because we couldn’t get a project's accuracies to fall below or anywhere close to 0.1ft.
How to Avoid It
To really rule out any issue in the control set, all our pilots must comply with 2 rules:
- Document with a photo the accuracy levels of the GPS at point collection
- Measure each GCP twice, once before the flight, once after the flight.
At AirWorks, we discard every point that has readings above 0.1 ft in the on-site photo documentation, and that has differences of more than 0.1 ft in any direction between the pre- and post-flight measurements. This way, we can ensure that a) our reading is correct, and that b) the point has not moved during the flight.
4. Poor flight layout and settings that yield blurry orthomosaics.
Why It’s a Problem
We see a lot of aerial data at AirWorks. Often, one of two things happen:
- There are blurry and distorted patches in the imagery in the middle of a project
- There are blurry and distorted patches on the outside of the flight path – especially on buildings and the building side facing the flight boundary
The second issue is more common than the first. Distortion in the imagery can be amplified when a survey is flown very high, which reduces the ground sampling distance.
How to Avoid It
Two factors to avoid blurry orthomosaics:
- Make sure that you plan your flight boundary at least 100 ft of the area that you’re interested in mapping. At the boundary the number of images used for matching is lower, so you need extra space to guarantee good results
- Make sure you have the right image overlap. We generally fly with at least 85/80 to ensure we can get enough data to model each object accurately. If the purpose is 3D modeling, in addition the camera angle should be adjusted to 30%.
5. Too much data for your computer or stitching software to process.
Why It’s a Problem
Large projects over 100 acres can produce more than 10,000 images with standard flight parameters. Those projects will typically cause off-the-shelf tools to crash--and if they don’t, it will take multiple days to complete the image stitching. Despite using some of the fastest computers here at AirWorks, we still couldn’t handle the amount of data from some of our large projects. In our case, we had to re-fly the project, but we did learn ways to prevent re-flying in the flight prep stages, which we’re able to share with you now!
How to Avoid It
Split the project up into various parts. Pix4D suggests splitting up your projects in sections of around 2000 photos, while we’ve been able to process projects with around 4,000 – 5,000 pictures without trouble.
When designing a flight with multiple sections, use enough ground control points along the section boundaries, or even overlap the sections. This will prevent that the last error mentioned above—meaning your images will have good overlap and boundary coverage, yielding crisp and clear orthomosaics.
There are enterprise tools available that stitch larger projects but they’re very expensive, so you’ll find it can really be worth it to split up the files and use cheaper tools.
6. Stitching issues and erroneous ground surface data from mapping in wooded areas.
Why It’s a Problem
Photogrammetry is the go-to solution for most projects--but really has its limits in wooded areas. Images don’t stitch well in heavily wooded areas because the algorithms can’t find matches between shifting tree tops. Another problem is that the top of the canopy is stitched at canopy elevation, providing erroneous data on ground surface.
In the end, the topography is inaccurate because these stitching issues introduce the following major sources of error:
- Ground elevation data is not available, which also causes gaps in the Digital Elevation Model (DEM);
- Contour lines are being drawn on tree top elevation.
This is problematic in both cases, as engineers need accurate data they can rely on. Imagine doing earth work calculations on elevations differences of 40 feet. Luckily these errors are often spotted, but if this inaccurate data were used for estimates, the discrepancies could total in millions of dollars.
How to Avoid It
Wooded areas are and will always be problematic for photogrammetry, as images cannot penetrate dense tree canopy. Most suggested solutions increase the likelihood of stitching images correctly and accurately calculating ground elevation, but they won’t eliminate the problem entirely. The only real solution is using lidar data collection- which more than quadruples the price of the project. Therefore, engineers often rely on photogrammetry results supplemented and verified by other means as a cost-effective compromise.
Our suggestions are:
- Fly higher in wooded areas. While a flight elevation of 150 ft leads to optimal ground sampling distance, often images in wooded areas don’t stitch at all. For wooded areas, we like to fly at 300 ft or 400 ft to make sure that we can stitch images more accurately. Due to the increased field of view, the algorithm can incorporate more of the surroundings, and therefore, is better at finding matches between different images. For the same reason, you also really don’t want to compromise image overlap here. We even recommend overshooting your image overlap in wooded areas.
- At AirWorks, we always highlight areas that don’t have topography or where the model is caught the tree canopy. In those areas, we verify and supplement our results with state available data from NOAA, for example, that has been collected with lidar. This way we can validate our results and provide engineers with accurate data.
6. Building footprints are drawn from top-down orthomosaics.
Why It’s a Problem
The footprint is enlarged and doesn’t account for any building overhang.
How to Avoid It
Always draft buildings in 3D.
In general, ground control makes or breaks an aerial mapping project. Being diligent at that stage will save you a lot of time and increase your accuracies.