Have you ever seen ghosts in your orthoimagery? Have you looked? Maybe you should… How about goblins crawling up out of your surface model, or gremlins messing up your point cloud?

The drone industry is filled with case studies of perfect projects and examples where everything has gone right. So, just in time for Halloween, we at GroundPoint Engineering thought it would be a great time to talk about what happens when the goblins come out to play…

Experienced drone teams know that sometimes you can end up with some spooky stuff! One true measure of a high performing team is not how well they do when everything is going right, but how well they perform in the face of adversity. I said in a previous article that drone operations for AEC should be boring. But even when you think you’ve done everything right, things can still come out sideways.

Knowing how to spot that something is wrong is half the battle. Knowing to go looking in the first place is something separate altogether! Our position at GroundPoint is to always take the promise of any “hands-off automation” with a healthy dose of skepticism…and perhaps a few grains of well-placed salt. Looking for (and finding) artifacts and errors can go a long way toward improving data accuracy, quality, and useability, even if at first glance things might appear to be “good enough”. It’s easy to get lulled in by the “wow factor” when lurking just below the surface are issues you didn’t even know were there.

The core data products from any drone mapping mission are the point cloud, the orthoimagery, and the surface (terrain) model. The orthoimagery is fundamentally an image mosaic made up of multiple overlapping images, but what makes it unique is the ability to produce accurate horizontal measurements. One basic assumption of an orthoimage is that it is a single, combined image representing a nadir view, or a view looking “straight down”. Because it is a mosaic compilation, “data ghosts” can occur when something shows up temporarily in one image but not in others overlapping the same spot, or because something in one image moved, so it shows up in another image in another place. Ghosts.

Ghosts can present challenges for software algorithms trying to resolve pixels between overlapping images when objects seem to appear and disappear! And “can presents challenges” is just a nice way of saying it reduces the accuracy of the final data. Ghosts can usually be resolved by reviewing the image catalog and selectively removing images that are creating the ghosts. In so doing, sometimes using FEWER images can actually produce BETTER data.

Unfortunately, there is no “instant ghost removal” button. That’s why it’s so important to look at the multiple data outputs and manage them thoughtfully. As much as we’d all love to think that any data collected is immediately usable “as-is”, (But wait! The hardware and software vendors promised!), the reality is that your data is going to need some focused attention during the post-processing or goblins are going to get the best of you.

Notice this nice suburban scene below. How many ghosts and goblins can you spot? At LEAST three! The car is an obvious ghost. There are gremlins messing with the powerline. And each backyard is full of gremlins!

How many ghosts can you spot in this image?

These kinds of gremlins (aka “data issues”) can occur because of the oblique nature of many images used by typical drone flights. The closer all the imagery is to being truly nadir (aka straight down) then the less likely this is to occur. When the algorithms are trying to reconcile the tops of objects from the sides of those same objects, while at the same time creating a seamless image mosaic covering the entire area, sometimes the sides of objects can get “smeared in” to the final image. Trees are an obvious example, but so are cars and buildings. This means that the accuracy of horizontal measurements extracted from these locations can be compromised.

Goblins can show up in the point cloud as artifacts or unwanted items that just need to be classified and filtered out of the current view. Sometimes the same goblins are the result of post processing trying to resolve data coming from two separate scans or two separate positions or angles. These can often be eliminated by improved positional data (aka better GPS), bore sighting, and some advanced point cloud manipulation. In these cases, an automated solution is probably not going to cut it.

Sometimes, other goblins show up in the ortho that are the result of shadows in the point cloud. This happens when oblique imagery does not provide a nadir look at the ground and there are “empty spaces” in the capture data that the software automatically fills in with imagery bits. For a lot of drone data, exploration of the point cloud can give some obvious answers as to why the ortho has some wonky bits, even when the overwhelming majority of the data is of very high quality.

Imagery and corresponding point cloud

Improvements in flight planning and flight operations are usually the best way to ensure that ghosts and goblins don’t take over your data products and they stay buried where they belong!

We at GroundPoint Engineering wish you all a very happy and safe Halloween, and hope that you keep your data ghosts at bay! Feel free to check out our previous LinkedIn posts for more on AEC surveying, inspection, and drone-assisted data collection.