Subscribe
Share
Share
Embed
The Drone Network explores how drones are reshaping the world. Hosted by Bryce Bladon, the podcast documents the tech, economics and people piloting the world's largest standardized drone imagery network.
# TDN214 - What Your Drone Actually Sees
**Speaker:** [00:00:00] A roofing contractor needs to estimate repairs on a commercial building. They've got drone footage, and it's beautiful footage actually. Smooth orbit around the building, nice light, clear detail on the roof surface. They try to measure the damaged section from the footage and they get the wrong answer.
And it's not a little wrong. It's, it's wrong enough to misprice the job, wrong enough that it matters. And the reason they got the wrong answer has nothing to do with the drone, nothing to do with the pilot, nothing to do with the camera either. [00:00:30] It has to do with a property of regular photographs that most people don't think about but that is fundamental to whether drone imagery is useful as data or just pretty to look at.
A photograph is a perspective projection. Objects closer to the lens appear larger. Objects at the edges are distorted relative to the center. If a building is tall, it doesn't appear straight. It appears to lean outward from the frame. If you try to measure distance between two points on a raw aerial photograph, you'll get the wrong answer because [00:01:00] the image isn't corrected for any of these distortions.
What the roofing contractor needed wasn't a photograph They needed an orthomosaic, and those are not the same thing. Welcome to The Drone Network, the only podcast in the air and on the airwaves. I'm your host, Bryce Bladon, and on this show, we document the tech, economics, and people piloting the world's largest standardized drone imagery network.
Today, I want to demystify what actually happens between the moment a [00:01:30] drone pilot closes the mission and the moment someone uses that data to make a decision, 'cause I think there's a gap, a real gap, between what most people think drone imagery is and what it actually becomes when it's processed, particularly when it's processed correctly, particularly when it's processed and used well.
And closing that gap and demystifying it matters for pilots, for clients, and for anyone trying to understand why a standardized aerial [00:02:00] data network produces something fundamentally different than a collection of drone videos. So we're gonna start with the photograph versus the data point, because listen, this is something basic, but it's not actually obvious.
When you take a photo with any camera, your phone, a DSLR, a drone, you're capturing a perspective projection of the world. Light enters the lens from every direction in the scene, and the sensor records where all that light lands. The result is an image that represents what the scene looks like from a specific point in space at a specific angle [00:02:30] through a specific lens.
That's a photograph. It's a record of appearance A data point is different. A data point is a measurement. It has coordinates. It has known accuracy. It can be compared to other data points collected at different times or by different instruments. It occupies a specific, verifiable location in the real world And when a drone pilot flies a standardized grid mission, a Spexigon, systematic parallel passes at a set altitude with calibrated overlap between frames, each image is [00:03:00] geotagged.
The drone’s GPS records its precise location in space at the moment of each shutter trigger. The metadata embedded in each image file contains that position along with the camera’s orientation. And this is, as funny as it might sound, where the interesting part starts because now you have not just photographs but photographs that know where they were taken, when they were taken.
And when you have a large collection of overlapping photographs and you know where they were taken, you can do something remarkable with them. This is where it gets interesting. The process that [00:03:30] turns a grid of overlapping drone photographs into a usable map is called photogrammetry. And the core of modern drone photogrammetry is an algorithm called Structure from Motion, SFM for short.
If you take a photograph of the same point in the real world from two different positions, you can calculate where that point actually is in, in 3D space using the geometry of the two positions and the apparent shift of the point between the two images. This is the same principle your [00:04:00] own eyes use for depth perception.
Your two eyes see the world from slightly different positions, and your brain uses the difference to infer depth. Drone photogrammetry does this at scale with hundreds of thousands of images instead of two. The software identifies common features across overlapping images: a rock, a root, a painted line on pavement, and tracks where each feature appears in each image.
From those matches, it calculates the position of the camera at each shot and reconstructs the [00:04:30] three-dimensional geometry of the scene. The result is called a point cloud, a dense set of three-dimensional coordinates, each one representing a measurable location in real space. That point cloud is then used to generate the deliverables that a lot of clients actually use: orthomosaics, which are geometrically corrected aerial maps where every pixel represents a true horizontal distance on the ground; digital elevation models, which are surface height maps; uh, 3D meshes, which are navigatable [00:05:00] models of the scene geometry.
The orthomosaic is the one I wanna focus on, though. So what is an orthomosaic and, and why does it matter? An orthomosaic is a high-resolution aerial image that has been corrected for lens distortion, for camera tilt, and for terrain displacement. Every pixel is repositioned to represent its true ground location, and the scale is uniform across the entire image.
You can measure distance, areas, and angles directly [00:05:30] from it the way you’d measure from a paper map. To understand why this matters, think about what you’re looking at in a raw aerial photograph versus an orthomosaic. In the raw photo, a tall building leans away from the center of the frame. In the orthomosaic, it doesn't.
The correction algorithm has repositioned every pixel to account for the perspective distortion. In the raw photo, a roof that was at the edge of the frame appears to have different dimensions than the same roof in the center. In the orthomosaic, it doesn't. The [00:06:00] scale is consistent across the whole image.
So for a client who needs to make decisions based on what they're looking at, this difference isn't cosmetic. It, it matters. A, a civil engineer calculating earthwork volumes needs accurate area measurements. An insurance underwriter assessing rooftop condition needs accurate area measurements. A municipality tracking impervious surface coverage needs accurate area measurements.
None of them can do that from raw drone footage, no matter how clear it is. The accuracy of a well-processed orthomosaic is typically stated in [00:06:30] terms of ground sample distance, GSD, which is the real world size represented by a single pixel in the output image. A drone flying at around one hundred and twenty meters altitude with a typical camera will produce an orthomosaic with a GSD of roughly three to five centimeters per pixel.
That means each pixel represents three to five centimeters of real ground. High precision workflows with RTK positioning and good ground control can achieve absolute horizontal accuracy of one to two centimeters. For reference, the best [00:07:00] commercially available satellite imagery has a resolution of around thirty centimeters per pixel, and that's the expensive, tasked, on-demand variety.
Uh, it's, it's not something you'll ever be able to get for free. If, if you were to get free satellite imagery, that is measured in meters per pixel. But drone-derived orthomosaics at three to five centimeters are operating at a fundamentally different level of detail. This gets to something that is probably worth making explicit, especially for pilots on the LayerDrone network using the Spexi app.
When you fly [00:07:30] a standardized mapping mission, you are not producing photography. You're capturing input for a data pipeline. The photographs are necessary, but they're not the product. The product is what they become after processing: georeferenced, geometrically correct, queryable spatial data. And this distinction is everything.
It, it's why the mission parameters matter so much, for one. Flying at the right altitude, maintaining the right image overlap, typically seventy to eighty percent between images and between flight lines, it's, it's the minimum requirement for the [00:08:00] photogrammetry to work correctly. Below a certain overlap threshold, software can't find enough common points between images to reconstruct this accurately, and individual frames are fine, but the data set they constitute isn't viable.
It's why standardization matters. One of the core things that makes a drone network valuable, as opposed to a collection of individual contractors producing one-off drone-related deliverables, is that the data is comparable across time and geography. If every pilot sets their own altitude, [00:08:30] their own overlap, their own camera settings, varying drone models, you end up with a collection of data of varying resolution, varying accuracy, varying metadata completeness.
You can look at it, might even look good- But you can't query it systematically, and thus it can't be the foundation for a system. You can't compare this week to last month. You can't run a consistent analysis across different cities. You can't do a lot of things that would make this matter. Standardized missions produce standardized data.
Standardized data is what can be indexed, queried, and turned into a kind of temporal [00:09:00] layer that makes fresh aerial imagery actually useful. So what I'm trying to get at here is explain what drone mapping does, and really at its core is it takes physical reality and it makes it legible to computational systems.
The world exists. Buildings stand, roads run, forests grow, coastlines erode. Most of that physical reality is not in any database. It's not queryable. It doesn't have timestamps. It can't be compared to what it was last month. It just exists somewhere between the map that was drawn years ago and the [00:09:30] ground truth that's changed since.
Drone mapping changes that, fundamentally improves that. A point cloud of a city block is that block in a form that engineering software can reason with. An orthomosaic of a flood-affected neighborhood is that neighborhood in a form that emergency response can act on. A time series of orthomosaics over a construction site is that site's history in a form that the project manager can audit.
Every photograph a drone takes is raw material. Every processed orthomosaic is a [00:10:00] piece of the physical world made legible, timestamped, georeferenced, queryable, comparable. The pilot who flies the mission probably won't see the GIS layer that their data eventually ends up in. The insurance underwriter probably won't know anything about structure from motion.
The city planner probably won't think about overlap ratios and ground sample distance. But all of those things, the algorithm, the overlap, the ground sample distance, the orthomosaic, are what make the connection between the drone in the air and the decision being made on the ground actually work. [00:10:30] And that's what your drone actually sees when you do it right.
Not photographs, infrastructure. That's this episode. Bit of a shorter one, but some big hard words for me to get my head around. I hope you enjoyed. I've been Bryce Bladon. This is The Drone Network. Thank you so much for listening. Just a little note at the end of this episode, I used a lot of sources for this one to make sure I got some details right here.
So I want to thank propelleraero.com, uh, [00:11:00] future3d.com, skybrowse.com, and commercialdroneguide.com. Uh, I verified a lot of my figures across those sites. Um, thank you again, and thank you for listening. Thanks for being a part of The Drone Network. Subscribe wherever fine podcasts are served to get a new episode every week.
And remember to leave us a five-star review on your podcast app of choice. It helps a lot. Today's show was sponsored by Spexi Geospatial and LayerDrone. Learn more about standardized drone imagery built for global scale at spexi.com. [00:11:30] That's S-P-E-X-I.com, and LayerDrone.org. Thanks again for listening.