Ted Strazimiri of SkyDeploy narrates "Infrastructure Megascans", a project video on capturing whole bridges at sub-millimetre resolution with drones and photogrammetry so that AI partners can segment and quantify cracks and other damage. Ted walks through a validation week and a year of capture on a three-kilometre Chicago tollway, then shows how the DJI Matrice 4 and Skydio X10 make the follow-up Blue Water Bridge project more efficient.
“But we're not after a beautiful model. We're after a model that we can extract the damage from.”
“we didn't really realize this until we tried it, which is why validation is so important before we start a project like this”
“You can't really just throw this into one big reality capture project and hit go.”
Yes. Ted Strazimiri used the Skydio S2 and later the Skydio X10, which navigate visually and fly fully autonomously in GPS-denied areas under the deck.
The Chicago asset owner asked for 0.6 mm GSD on every surface; according to Ted Strazimiri, the car-mounted rig reached 0.1 mm and the Skydio X10 can reach 0.3 mm, though image counts grow exponentially.
For Ted Strazimiri's project it was still ongoing at the time of the talk; the data has to be processed and delivered in chunks rather than as one big project.
Transcribed automatically (ElevenLabs Scribe) and lightly edited. It may contain small errors. Click a time to play the video from there.
Ted Strazimiri: Thank you. Hi, everybody. So nice to be in a country where everyone can say my last name correctly. It's been a while since I've experienced this. Uh, lovely to be here.
Ted Strazimiri: Thank you everyone for organizing this, and I have a short, well, twenty-minute video to take you guys through. I'm gonna be narrating through parts of this, and, uh, it's a, uh, pretty interesting project that I've been working on for a while, but I haven't really been able to share too much until now. Uh, so you guys are ready back there to play the video? Cool. So I'm calling it Infrastructure Megascans, and I started out by using drones, specifically around twenty seventeen.
Ted Strazimiri: I fell in love with photogrammetry, and everything that I've done has been self-taught up to this point. I stumbled across several use cases, including parking analytics, which I ended up starting a second company called Parklytics, uh, for. And predominantly, my day-to-day revolves around SkyDeploy. Um, that's the business where I provide services, reality capture servi- services, drone operation services, et cetera. It leads me to come across a lot of different, uh, applications and, and use cases, and I like to make videos, and really make the videos for myself, um, kind of to have a portfolio of my work.
Ted Strazimiri: This video probably is the reason why I'm here today. This video was, uh, about three years ago, two to three years ago. Um, and it was also one of the videos that probably took me the quickest amount of time to just put together and, and post, but it's also the video that did the best in terms of impressions, and I was blown away by how interested people were in this concept. What you're seeing is the segmentation of infrastructure damage, specifically cracking efflorescence, um, and, uh, that's been overlaid onto the three-D model. It's a really impressive, uh, visualization tool, but also, uh, what's really impressive about it is that it provides the quantities of damage as well.
Ted Strazimiri: This was done by a company called Aren, a small startup based out of New York. Uh, I ran into them, and I was really, uh, really blown away by, by their attention to detail and also how many models they had already set up and prepared, uh, to perform this type of inference. They take in any kind of data, but it works really, really well, of course, with high-quality photogrammetry data. Here's a airport runway, uh, where we, uh, scanned. Actually, this was a very interesting project, so we scanned this while the airport was open, and then we, uh, sent that data over to Aren, and they performed the analytics.
Ted Strazimiri: Of course, AI behind it all, and, uh, extracting those segmentations of the crack damage. The most important factor is typically the crack width, and so that was something that if we had to do this by hand would have taken forever. That video, along with, uh, a, a few more of my LinkedIn posts, led to this project. This is a, uh, tollway in Chicago, and in twenty twenty-three, the asset owner reached out to Aren, and I was involved as well in this, uh, to begin a scan of the entire structure. We wanted to achieve essentially the same results that you saw in that, uh, previous bridge video, but we had to do it across three kilometers.
Ted Strazimiri: The kicker is that they wanted every surface done. The wearing surface, the barrier walls underneath the bridge, everything, and they wanted it done at point six millimeter GSD. That's where, uh, I got a little bit nervous, 'cause that is, uh, uh, an insane amount of data that needs to be collected for all of this. And the intricacies of these geometries under- underneath here makes this very challenging. Difficult lighting conditions.
Ted Strazimiri: But we took it on. We, we put through a proposal, and they said, "Okay, show us what you got." So that started with our first week. We call this the validation week. We had a lot of assumptions to test out.
Ted Strazimiri: So we took a week to pick certain locations within the structure or across the structure and validate those assumptions. What drones, what cameras, what techniques were we going to be using? We started by collecting a very large-scale, uh, three-D model, loaded that into Nira so that we had it available at field's edge. This was very useful because often we need to check clearances and heights and distances between points, and we only really had one week to get all of this done. And a lot of this required significant lane closures.
Ted Strazimiri: You'll see, uh, this is Ali, the CEO of Aren, and he's putting down some forensic scale tape. This high bridge component, what you're seeing here, the highest part of the structure, was the most challenging, so we decided to start with this first. And what's very interesting, and we didn't really realize this until we tried it, which is why validation is so important before we start a project like this, is that inside of this area, that structure overhead essentially works like a Faraday cage. GPS is very much compromised in there. So we really resort to a very unique drone, a Skydio S2 This drone is able to operate in a GPS-denied environment, and it's fully autonomous.
Ted Strazimiri: So the operator just simply gives it a scanning volume, and then the drone goes and plan- plans its path around that volume and the surfaces within it to capture the correct overlap and correct number of images for that photogrammetry scan to be successful. This was definitely the most intense situation that I had been involved in with traffic and the bridge remaining open. Of course, this is a tollway and connects very key, uh, sections of Chicago. And there were some areas where we needed to get a little bit creative. So I hacked together this little contraption.
Ted Strazimiri: This is a Fuji GFX100S with an Emlid M2 RTK up top. And this was done for the barrier walls, as you'll see. There really was no way to collect the burial-- barrier wall imagery on the inside with a drone. And we had three kilometers, but six set of-- six sets of walls to collect. This is the setup, my friend Seth's car, and you can see very rudimentary, just getting it done.
Ted Strazimiri: But it worked. Having PPK tagged images on import really helped with the alignment, especially for long strips, especially we didn't really have the opportunity to do too many passes, just one pass forwards and one pass backwards of images. This resolved down to point one millimeter, which was better than we had promised. But of course, we couldn't maintain that point one millimeter across the entire bridge. So in most cases, we were on average about point six millimeters.
Ted Strazimiri: Another part of the test week, here's a-a-another difficult element of this bridge. And lucky for us, they had just painted this bridge this beautiful brown color about a year before we arrived. And of course, you may definitely have experience with, uh, uh, you know, capturing for photogrammetry, and, uh, you know how difficult that color is to work with. Of, of course, it's also s- in some cases reflective. So it was a really big, um, uh, you know, pain for us.
Ted Strazimiri: But we're not after a beautiful model. We're after a model that we can extract the damage from. You can see the S2, the Skydio S2, that's the closest distance it's able to fly, uh, to the surface, and that's what we had to use, uh, in order to g-- achieve that GSD, that resolution. But that also meant that the, uh, capture time, uh, grew exponentially. Some other difficult, uh, areas.
Ted Strazimiri: S-- uh, close quarters, uh, low light in these, uh, locations as well. And you see a lot of Skydio in this, in this video, but we have M300, we have DJI drones. We're throwing everything we can at this problem on that week to find the sweet spot, the right combination of tools. It's always about having the right tools for the job. You can turn the volume up a little bit in the back.
Ted Strazimiri: And this drone f- completely flies itself, so there's no pilot operation in these stages. We even went as far as to test bringing out a floodlight. Um, in some areas, days got short, and we needed to get things done. A few weeks later, after we were happy with everything, we returned and started collecting. And this went on for a long time.
Ted Strazimiri: See, we had two lanes closed at a time. Drivers were not happy with us. They couldn't really tell we were working, so I got a, a lot of, "Hey, get back to work," yells out the window. I'm working. We did have to utilize Starlink on a few occasions because the cellular connectivity down there was so poor, and we had to get the data over to post-processing, uh, as quickly as possible just of-- because of how much data we had to go through.
Ted Strazimiri: So 12 months later, the project wrapped up. Over five hundred terabytes of images collected, and there's one NDA still between me and being able to show you the results. But that's okay because there's another project that's come up recently that I am able to show you a lot more of. And there's two new tools, two new drones that we're able to use, which really, uh, has a huge impact on efficiency. The first one being the Matrice 4.
Ted Strazimiri: I had the ple- the privilege of visiting, uh, DJI in China in November. Got to see the, uh, M4, uh, in person and bring it back with me, and right away, I knew this was gonna be a game changer for us in these scenarios. The M4 solved a lot of problems we had with path planning, and creating these missions ahead of time was very time-consuming, and often when we arrived on site, we saw misalignments between the p- the path for the drone that we had planned in software and then what the drone was doing in r- in, in, in actuality. So having that ability to do it in real time, uh, really makes things a lot easier for the field work. And the second drone that really was-- is game changing for this use case is the Skydio X10, the new generation of the S2, which you saw earlier.
Ted Strazimiri: This drone now has fifty megapixels instead of twelve, a much better navigation system, more flight time, and we can achieve resolutions of point three millimeters if we want. Not saying that you have to, but if you want. Again, the number of images grows exponentially, so post-processing remains the bottleneck, but we can do some fun things. The other aspect of this drone is that it has co-aligned RGB and thermal cameras. So as we're flying these bridge inspections, we're not only just getting RGB data, we're also getting radiometric thermal IR data, which is exceptionally good for inspections.
Ted Strazimiri: And predominantly, a lot of my inspections are actually video-based. Um, it's only when we really need to extract quantities that we resort to photogrammetry, but having both there is really useful. Again, completely GPS denied, which is wonderful for us to apply to the new bridge. This is the Blue Water Bridge, right at the border of US and Canada. This project kicked off, uh, about a month ago, and it's going quite well.
Ted Strazimiri: And there's my friend Jeremy, who's probably in the audience right now. Hi, Jeremy. So the X10 is really game changing. You saw an upward capture mode. That upward capture mode allows us to com- use the- these two tools together.
Ted Strazimiri: So the Matrice 4 in the slope capture for everything on the exterior above the deck, and even in some cases below deck. And that's predominantly I've been using slope capture, which is again, this real time setup. You're doing this in the air on your screen, dragging out this area, setting up the overlap and the scan direction. And this is based off of RTK nav-navigation, so this is persistent. You can go back between multiple days, multiple batteries.
Ted Strazimiri: And it's very foolproof, even in areas of what you would expect high multipath and, and poor GNSS conditions. I've experienced that it works very well because the M4 now has a hybrid approach to its navigation. It's using its visual odometry in those navigation cameras, as well as GPS to supplement it. And when-- you can go a, a quite significant time without an RTK fix and still retain good, uh, positional integrity here. But what you can't do is go under that bridge and point the camera up, and that's where the X10 comes in.
Ted Strazimiri: So again, just like Skyway in Chicago, a base map, a base model. This is very useful for planning. But the base model doesn't give us any, uh, in-- good idea of what's going on under the deck. That's where the Matrice 4 with that slope scan mission type comes in. It's quite good, but not great for these areas under the deck.
Ted Strazimiri: And that's where X10 supports. So this is M4, and then this is X10. So using this layered approach allows us to have everything and pick the areas where we wanna focus on. Makes the data collection much more efficient and much more redundant. Of course, everything you're seeing here is being, uh, rendered on Nira, my-- one of my favorite applications of all time, web applications.
Ted Strazimiri: These, these models are huge, and so being able to share them with our team, uh, in the field and, uh, clients is really important. It's game-changing to be able to do this in real time, and specifically at field's edge. Now, Cone Labs is a startup based out of Canada, quite local to myself, and they're doing something very similar to Aren, just in their own special way. This is the, uh, AI partner that, uh, is working with me on this capture, and here's a look at the results. So the segmentation, quantification, and visualization of this damage is very, very important, and it makes, uh-- it has a lot of value to the asset owner.
Ted Strazimiri: That awareness of where everything is and what they need to plan for, uh, in their very busy maintenance cycle. So I'm really looking forward to seeing the progress of this, uh, this company and many more, and I know there's others like it here today. So thank you. Uh, I really wanna thank, uh, everyone for setting this up. Michal, Tomas, thank you so much for having me.
Ted Strazimiri: Aren for allowing me to show this, and Cone Labs for working with me on this, uh, Bluewater project. And as well, Nira, of course. Couldn't do any of this without that tool, and that team is incredible. Um, it's really changed my business. Um, and I'm really happy to announce today that I'm joining Nira, uh, and joining that amazing, uh, but small team and helping them move, uh, this forward.
Ted Strazimiri: Thank you.
Host: Ladies and gentlemen, let's give it up for Ted. Well done.
Ted Strazimiri: Thank you.
Host: Well done. We've got actually... Ted, stay with me on stage.
Ted Strazimiri: Yep.
Host: Brilliant job with sending the, the questions. Actually, you sent over forty questions in, so we're going to take two, all right?
Ted Strazimiri: Yep.
Host: Just to keep this fast-paced. So Ted, uh, did the Skydio capture the wow steelwork structure with no gaps or holes, or did you have to manually fly some areas?
Ted Strazimiri: Great question. Um, you're referring to the reconstruction. There's tons of holes in the metal elements, but that's okay. Uh, again, we're not going for a pretty model. We're going for, you know, a model and the base photos that we can extract the damage quantities from.
Ted Strazimiri: The, uh, the Skydio works extremely well in all conditions. It doesn't have a magnetometer, doesn't have GPS, doesn't need it. It's all visual navigation. So for the Skydio, one hundred percent autonomous in flight and, uh, uh, execution. With some of the DJI drones, of course, manual flights.
Ted Strazimiri: Um, especially with the, uh, the, the Skyway project in twenty twenty-three.
Host: Thanks so much.
Ted Strazimiri: Yep.
Host: One more question. How long did it take to post-process?
Ted Strazimiri: Still ongoing.
Host: Still ongoing.
Ted Strazimiri: Still ongoing, yeah. This has to be done in pieces. You can't really just throw this into one big reality capture project and hit go. Uh, so breaking this up into chunks and working at it one step at a time, and of course, delivering the information one step at a time is also-
Host: Brilliant
Ted Strazimiri: the way the client typically likes it.
Host: Ted, thank you so much.
Ted Strazimiri: Yeah.
Host: Ladies and gentlemen, give it up for Ted one more time.
Ted Strazimiri: Thank you.
Host: Thanks. Perfect.

Drones & Data
Extremely agile and self aware individual. Subject matter expert in drone tech & photogrammetry. I have a deep passion and understanding of complex geospatial technology, business development, aviation and scaled sUAS operations.