How do drone inspections achieve accuracy?

[00:00:02] Two things, one was accuracy and precision. So bringing our models into the real world and making sure they're real world accurate. But then the other thing is. How we achieve accuracy? So in photogrammetry there is a thing called GSD, ground sampling distance. Ground sampling distance is what 1 pixel is in the real world.

So by way of example, if you were to look at some rural place in Ireland. Google Earth's GSD ground sampling distance might be 20 centimetres. That means one pixel on Google represents 20 centimetres squared of land in the real world, right? Okay, yeah. If you were to go to the middle of Dublin city where they have a much higher resolution of Google satellite, um. Might be looking at a GSD of maybe 5 centimetres. So that means one pixel in Google represents 5 centimetres in the real world. With drones, we can easily shoot at well below 1 centimetre GSD.

We are currently doing a project right now where our GSD level is 6 mm. That means one pixel in our data is millimetres in the real world. For reference, that's every blade of grass can be seen. Every pebble. Yeah, a hairline crack is in concrete is easily visible. That's the sort of GSD.

Now, do you want that level of detail? Not all the time. And one of the things it's very important to understand is, the good enough level of detail. The reason being the higher the GSD, the more data you're dealing with, the higher the cost of processing and so on and so on. So, you never want all the pixels, you only want good enough for what you're trying to achieve.

But pixels matter. Or GSD matters and the reason is there's a simple rule that says three times your GSD is the maximum level of accuracy you can hope to achieve for any given photogrammetric product. Okay, if I have a one centimetre level of accuracy GSD, then the maximum level of accuracy I can hope to achieve is probably about 3 centimetres. Okay. We always want to keep that in mind.

For that reason, we typically shoot between one and three centimetre GSD. For low accuracy stuff that the client doesn't need the high level of accuracy and doesn't want to spend a lot of money, we use up to three centimetres for very high accuracy, usually very small projects, maybe two or three hectares. Very high accuracy, almost always hardscape as well we'll shoot much higher. A much lower GSD which is a higher level of resolution. So down to as I said, the project we're currently doing has a six millimetre level which is very high relatively speaking. We could go higher. It doesn't add anything to the world really.

[00:03:13] Interviewer: Yeah, for asset inspections then do you typically use the higher because you really want to get in and look at like every brick or um. is there times when you don't need that accuracy on assets?

Well, okay, so there's two types of asset inspection. There's, the cheap one and the expensive one. Cheap one is I take a drone and I take a lot of pictures and I look for defects and I take pictures and then I show them to the client. They say, yeah, yeah, that's great. Thank you. That is more than sufficient for 99.99% of assets inspections. And an asset here can be a bridge, it can be a roof, it can be an industrial tower, it can be a water tower, it can be an entire farm, whatever. Okay, yeah, that's asset inspections.

I can also as an extended level of deliverable add in a 3D model of the same asset. Depending on the type of asset, we would certainly shoot it very high. So we recently did a project where we did a bunch of causeways in the West of Ireland. Those were all shot I think at 5 millimetre GSD, hairline cracks you can see all the features of each individual rock and these are all old stone causeways built of old granite that, you can really get a good feel for where the rocks are, they still have their integrity and where it is starting to crumble and fall apart and where some work is needed to be done.

Interviewer: In the Nira video you can fly under the bridge in the interactive 3D video.

It takes skill to photogrammetrically generate that because right, everything, everything that exists within a model must be seen by the drone. Yes, Yeah. So if I was to stay only on top and shoot down. Anything that's not visible, does not appear in the model it appears as a big BLOB or just nothingness. Whereas to get that so we we intentionally make sure our models capture all of the relevant data to perform an inspection and certainly in that particular instance the underneath of a bridge will be a very important part of what the client is looking to inspect.

[00:05:24] Another aspect that we do that almost every drone operator worth their salt is doing, which is they're using a system called NTRIP or RTK, real time kinematic is what RTK means. What that means is that there's a super duper hyper precise GPS built into the drone and it is receiving local corrections from ground stations as well. So instead of the drone thinking it's within let's say 50 or 100 centimetre bubble. We're now bringing that's that down really, really, really small. The drone now believes it's within maybe a two or three centimetre bubble.

A drones GPS antennas, need to know how far horizontally and vertically they are removed from the plane of the camera or the sensor. When we when we fly with LiDAR or photogrammetry okay, but it needs to know exactly the horizontal and vertical difference between them. The reason that's built in, the reason we utilise that is because that allows either the LiDAR or the photogrammetric data to be tagged incredibly precisely. Because of that incredibly precise tagging, you're able to get a much more accurate raw data set before ever bringing in ground control.

And one of the things I suggest to some of my clients is do not worry about ground control for this particular type of project, it'll cost you way more. Because ground control represents on average, three times the amount of work that the drone flying itself represents. Okay, so for every hour I might be performing LiDAR flying, I'm at least doing 3, probably 5 hours of ground control. Okay, because ground control walking around with a pole around the land takes a lot longer than the actual flight itself. It does, it really does. And so a major cost saving can be had if your project is worthy of it. That is to say, if you do not need a high level of accuracy.

You can get away with what's called RTK only flying. So you fly the whole mission with RTK and you never utilise ground control, and you have a lovely 3D model but you do not know if it is accurate or precise. Uh, yeah, it'll be pretty good, but there might be a little bit of variation and you don't have. You do you do not know if it is good enough for survey grade, but certainly if somebody was looking at just I need a broad pretty picture of my farm. I just need to roughly know what this asset looks like in high resolution, but I don't need to geographic, geolocate it within the world. That's fine. That is a quick and fast way of saving a lot of money while still getting the product that you need.