Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Photogrammetry - Episode 11
230 Plays10 years ago
On today's episode, Russell interviews Leszek Pawlowicz, a southwestern archaeologist. The talk about Leszek's work with different photogrammetry software and techniques. There is some good stuff in here about how to start in photogrammetry and it doesn't take expensive software or equipment.
Recommended
Transcript
Introduction by Host Russell Helene Willems
00:00:01
Speaker
You are listening to the Archaeology Podcast Network. Hello and welcome to the Archaeotech Podcast, episode number 10.
Interview with Leszek Kavlovich: Photogrammetry & Imaging Techniques
00:00:12
Speaker
I'm your host, Russell Helene Willems, and today on the show, I'm talking with Leszek Kavlovich, an archaeology technologist from the American Southwest. We're going to be talking about his work with photogrammetry and reflectance transformance imaging, or RTI.
00:00:33
Speaker
Welcome to the show, Leshek. Thanks for having me, Russell. So, Leshek, can you tell us a little bit about how you got started experimenting with photogrammetry in archaeology?
Early Experiences with Microsoft Photosynth
00:00:42
Speaker
It actually started seven years ago when Microsoft put up a website called Photosynth.net. And what that is, is you take a set of photos of an object or an area and upload it to the site.
00:01:01
Speaker
and it creates what's called a photo cloud, where it figures out what angle you were shooting all the photos at, generates a very simple 3D model, and then allows you to view the photos at the angles at which they were taken at.
Practical Application at Cordis Junction Dig
00:01:19
Speaker
I can try and bring up an example of that. That would be great.
00:01:23
Speaker
We're experimenting with the screen share feature here, so for listeners of the audio podcast, there'll be some extra content that we'll have as a YouTube video that you can watch as well. Okay, so website is photosynth.net, and I was doing a dig at the time at Cordis Junction in Arizona, a fairly substantial pit house village, and I took a fairly reasonable set of photos. I took about 130 photos of a pit house that we were excavating.
00:01:52
Speaker
And uploaded them to photosim. So this is a photo cloud.
Understanding Photogrammetry Software
00:01:58
Speaker
And as I move my cursor to various locations, you can see little squares those correspond to photos that I took And if I click on one of those equals zoom in on that particular area and you can zoom in closer Or zoom further away.
00:02:15
Speaker
And what it's actually doing is it's referenced all these photos to what's called a point cloud, which is a 3D representation of that particular model. So this is a very, very rudimentary form of photogrammetry, but it was the best that we could do back in six or seven years ago.
Creating 3D Models on Personal Computers
00:02:36
Speaker
And what all the photogrammetry software does is it takes multiple photos
00:02:43
Speaker
and figures out using perspective and multiple angles where the camera was located relative to all the features and then uses that information to generate a full 3D model. Now, more recently there are services and it's also software that allows you to actually create full models on your own computer or using a web service. And I've done that with the set of photos that I took here.
00:03:12
Speaker
And let me bring that up. OK. So this is an excavation in central Arizona. And basically, the digging's been done. And I was commissioned to go in and do a 3D model of the Pueblo that they excavated. And all this was generated using photos from ground level. There's no aerial stuff involved. And we zoom in a little bit closer.
00:03:40
Speaker
You basically have the freedom go in and rotate the model. Tilted And also you can generate a virtual overhead orthographic view is if you're basically looking straight down.
00:03:56
Speaker
on the model. And again, you took all those photos at ground level. You didn't have poles or balloons or drones or anything. Nothing like that. I had a ladder to help me get a little bit better elevation, but actually that really wasn't necessary. I've done perfectly acceptable 3D models using just ground level stuff with no ladders. And the software I'm using is a program called Photoscan.
Cost and Discounts for Photoscan Software
00:04:21
Speaker
And it's generally accepted as the best possible software currently available to do this kind of work. There's a number of different options, but I've tried some of them in demo form and they haven't worked that well. Photoscan is really nice. The downside of Photoscan is that it's reasonably expensive. The basic version costs $179 and the more advanced version that's used for
00:04:45
Speaker
aerial 3D models and for geographic calibration is about $3,000. However, if you're a member of an academic community or if you're a nonprofit, you can get a substantially discounted version of both of those. I think it's $79 for the basic version and $500 for the advanced version. If you just want to play with
00:05:10
Speaker
with the general concept of photogrammetry, there are a number of free programs and free services out there that will let you do that.
Challenges with Free Photogrammetry Services
00:05:18
Speaker
Yeah, I know, for instance, AutoCAD has their, what used to be Photofly, and I think it's now 123D cache. And there's the differences with that. I think that when you actually upload your photos to their server. Yeah, well, doing the processing for photogrammetry is an extremely, extremely time
00:05:39
Speaker
and computer intensive effort. Really fast computer about four gigahertz quad core eight thread with 32 gigabytes of RAM and running the last model I did that I showed you took about 24 hours of continuous processing time. So that is not messing around. Yeah, it's not definitely not messing around. A service like 123D catch is nice because all the computers at Autodesk
00:06:08
Speaker
handle all the basic processing stuff. There are some drawbacks, though, with 123D Catch. The biggest one is that you're limited to only 75 photographs, which means that you're also limited in terms of the total resolution that you're capable of getting. The model I showed you had five million mesh faces, and it's actually downsized from the original one. The maximum number of little mesh faces you can have with 123D Catch is 500,000.
00:06:37
Speaker
There's a free version of 123D Catch that has the limitation that anything you put up there is both publicly visible and is also the property of Autodesk. They can use it for any purpose. I believe they've got a private pay plan that will let you sort of keep stuff private and let you use it any way you want. There's also limitations on commercial uses as well too. I've got actually a couple of lists of programs as well too. There is a Linux distribution
00:07:08
Speaker
ARCH, which has something called the Python photogrammetry toolkit, which is a sort of a simplified graphical user interface that allows you to create 3D models. The software that uses is a little bit buggy and sometimes doesn't really produce optimal results, but it is something that you can try. On the Windows side, there's a program called Airphoto SE
00:07:34
Speaker
And I'm not going to give links to these. You can basically just go ahead and Google air photo SC one word and It is designed primarily to take aerial photographs and sticks them together into an ortho photograph and also create a digital terrain model, but you can also use it in general to create 3D models of of of an architectural feature or an object or just a general area.
00:07:59
Speaker
So choosing some of the same principles, you know, taking photographs and kind of finding where the angles match up, just it's maybe designed for more of an orthographic top down view. Yes, it is. But it will generate a 3D model. And I've gotten reasonable results from that. It's sort of like an overarching program that takes other programs and kind of creates this Frankenstein monster that's able to process this stuff and create 3D models.
Autodesk's Recap & Creating Detailed Models
00:08:26
Speaker
There is a
00:08:28
Speaker
If you're in a nonprofit or if you're in a university institution, Autodesk actually makes a free advanced version of 123D catch called a recap and recap 360 that is good not just for generating 3D models, but also is good for looking at and analyzing LiDAR data. Very cool. So you can kind of use different sensor packages with that one. Yep.
00:08:52
Speaker
And there's also a couple of web services in addition to 3D catch that allow you to upload stuff and create more detailed models. One of them is called Arc3D, one word. And the other one is CMPMVS web service. And the idea with these is you sign up for the service, you upload your photos, they've got computers that do all the number crunching, and then anywhere from a few minutes to a few hours later, you get an email saying, hey, you can now download your
00:09:22
Speaker
of your model, and then you could load into software on your computer and play around with it a little bit more. So with all these different software packages, do you still have the same or similar process for creating the photographs that go in to these models?
Photo Requirements for 3D Models
00:09:36
Speaker
And about how many photographs, and you said there's five million mesh faces in the model that you just showed us, how many photographs does that equate to and about how long did that process take?
00:09:46
Speaker
Okay, the 5 million mesh faces is actually downsized from the original, which was 65 million mesh faces. And that's really pushing the limits of what most computers can handle. I mean, it will display and turn very, very slowly. The number of mesh faces is going to depend on a number of parameters. It will depend partially on the number of photographs you take. And that in turn will be a function of the total amount of area that you want to cover in your photogrammetric model. Or a small
00:10:15
Speaker
feature or a small artifact, you could easily get away with only 15 or 20 photographs and get a reasonable model. For the model that I showed, fairly large area, fairly extensive, I wound up taking about 4000 photographs and wound up using about 1600 of them in generating the model. And that's one of the reasons it takes so long because the more photos it takes, the more time the computer takes to figure out the matching points in the photographs, what angles were taken, and then generating the full 3D model.
00:10:45
Speaker
So kind of different resolutions for different uses and that's interesting that it only takes about 15 to 20 for a small artifact. Whenever you're out in the field, how do you make a decision about when you've done enough or do you typically shoot for as much coverage as you can? You can always, my philosophy is shoot as many photographs as you possibly can because electrons are cheap.
00:11:08
Speaker
and time is valuable. So if I take too many photographs, I can always chop out some. If I take too few, it means I'll have to go back. And in fact, on this, the model that I showed, I went, I shot, I think, 3,000 photographs and then decided now there's a couple areas where I'm missing coverage and had to go back and shoot a few thousand more. And then from those kind of
00:11:31
Speaker
come up with a set that covered everything and gave me the resolution that I was really looking for. And one thing I noticed about your model is I didn't see any people in the photographs. Was that intentional or was that a function of did you have everybody clear out from areas you were photographing and then stitch those all together? On the first set of photographs I had somebody helping me and the software basically relies on what your photographing could be standing still.
00:11:57
Speaker
So if you have people that are moving around, it will generally be able to not be able to say, OK, I've got multiple angles of that person because they've moved from one location to another and not include those. However, sometimes I've had a couple of models where somebody stood in a particular location a little bit longer than I would have liked them to, and they'll show up. But you can always load the model into an editor and chop them out so that doesn't distract them. So if you have a disembodied hand from an excavator sitting in one of the frames,
00:12:25
Speaker
And it ends up you know that they were there for a little while you can then chop that part of the model out and fill it back in from other other photographs It you can It's a mix I mean sometimes if you chop out the actual 3d model and then you go off and you do a little bit of Photoshop work to sort of you know fill in the spaces where a disembodied
00:12:52
Speaker
body might be lying on top of the surface to sort of clean it up. And you can also do things like if you've got a dirt pile that you don't want to have in your model because it's really just, you know, backfill dirt that wasn't moved before you shot it, you can, you know, go in and edit it and sort of squish it down to the correct level. So it's partially letting the computer kind of algorithm figure out what to interpolate and then having a human go in and correct the model and kind of massage a little bit.
00:13:24
Speaker
Hey everyone, I'm back with Jordan Harbinger from the Art of Charm. Jordan, we've been growing our listener base over the last few months. Why don't you tell everyone again what they can get out of listening to the Art of Charm podcasts.
00:13:33
Speaker
Hey, so what we do at the Art of Charm, especially on the show, is we take tools that ultra high performers use and we make them accessible and we teach them to you. So what we primarily specialize in is relationship building, which is powerful for people in your field because any, well, any academic field or any field period, because as everyone knows, you only, it's all about who you know, right? And most people say that like, well, it's all about who you know. And they don't like that.
00:13:59
Speaker
because they're on the losing side of that equation. What we wanna do is put you on the winning side of that equation where you say, wow, I'm glad it's all about who you know because maybe this isn't the strongest or maybe I don't wanna sleep under my desk and try to outwork everybody, or maybe I'm doing both of those things, but I still want an edge. Your relationships are what's gonna deliver that. So we teach people how to do that, especially people who have kind of an analytical mindset, like a lot of people you might know, if you know what I mean.
00:14:23
Speaker
and they can apply these very practical skills. We don't say things like, just put yourself out there. I don't do that. I say, all right, what you're going to do is this and then you're going to get this information. You're going to connect these people together. You're going to follow up in this amount of time and you're going to say this. And that makes it a heck of a lot easier because there's no guesswork and there's no fingers crossed. Hope this works type of system. It's a real system. Awesome. Check out the Art of Charm podcast, wherever you get podcasts and at www.artofcharm.com.
00:14:59
Speaker
Oh, what are these models typically used
Recreating Excavations with 3D Models
00:15:01
Speaker
for? Is it still such a new area that people are figuring out what the utility is? Well, I wish Doug was able to make this podcast. He just did a blog post on some photogrammetric work they did at Walnut Canyon, where they did
00:15:18
Speaker
There were some areas where they could bring a LiDAR scanner in and to get a LiDAR model of a clip drawing, but there were some areas where they didn't feel like it was safe to haul a $200,000 LiDAR scanner. So they did photometric models and then they loaded them into AutoCAD
00:15:34
Speaker
And used auto CAD to basically trace out every single rock in a stone wall and all the features and also have this calibrated. So basically, you had a 3D, you know, drawing of that particular model. Another way that it's being used is for people who excavate
00:15:55
Speaker
Say, for example, if you're digging down a pit house and it's a really complicated structure with multiple floors and overlapping levels. You can basically excavate one level do a 3D model go down another level do another model and then when once all that's done, you'll be able to go back and recreate the whole process of the excavation and see what features appear and disappear as you go down to different levels.
00:16:19
Speaker
So that's interesting. You can actually use these photographs and the resulting models to recreate kind of the process of excavation and kind of the decisions that were made. I mean, you know, you certainly take photographs and you make drawings as you're going along, but there's almost no substitute for having an actual 3D representation that you can take a look at and rotate and see different
00:16:40
Speaker
Aspects of so whenever you take the time to build these models What do can they be viewed in on the other hand? Is it something like a 3d PDF or I know Doug Gann who you mentioned before and Rachel Opus Opus from cast have experiment with using game engines like the unity game engine to view it like in a 3d web browser and even to walk around in some of these areas virtually and
00:17:04
Speaker
Yeah, that that's something that looks really interesting and really exciting. I haven't actually played around with it yet. It's a little bit. It'll take some time for me to figure it out. I mainly use Many of these programs have a built in view or for example photo scan has the ability to view the model. There are also a number of free programs out there that will allow you to take a look at and manipulate models.
Editing 3D Models Post-Processing
00:17:28
Speaker
The one I use more than any other one is one called cloud compare
00:17:33
Speaker
And there's also, which is a very, very sophisticated program with lots of features, but also has a separate viewer program called CC viewer very high speed allows you to spin models and actually go into like a walk around the mode and a look around mode to examine them for editing and doing other work. There's a program called mesh lab.
00:17:54
Speaker
And another one called Meshmixer from Autodesk, which is not just a viewer, but is also an editor that lets you chop things out and smoosh things down and clean up a mesh that you've generated. So on the back end, whenever you're cleaning up these models and these meshes, what does the time budget look like in comparison to actually taking the photographs? Is it more on the back end correcting the model or more on the front end actually collecting the data for the model?
00:18:20
Speaker
Depends on how much work you have to do on the model. I did a number of artifact models for the Museum of Northern Arizona a while back, and some of them came out very, very nicely. Other ones had a lot of cleanup work to do. If there are shiny reflective surfaces on what is actually in your photograph of those that generate artifacts, and for those you really need to go in and sort of like swish the bumps down and smooth off the surface to make it look good.
00:18:50
Speaker
Other times I've generated models that look absolutely perfect the way they come out and require no more work other than maybe editing out a little bit of superfluous information. So it sounds like there part of the technique is understanding when you're out in the field taking photographs what the impact that's going to be later such as the lighting condition or whether you're taking pictures of reflective objects. Are there any tips that you kind of use in the back of your mind when you're taking photographs to compose shots or to make sure that there's always a different landmark in view in the back of the photograph?
00:19:19
Speaker
Depends on the software. Photoscan, most programs require that you have items visible in the background, and they use that as kind of a reference to help orient the camera positions. Photoscan is a little bit better than those in that you don't need to necessarily have as much background information. For example, when shooting a small artifact, I like to do it by putting it on a turntable and rotating it, snapping photographs.
00:19:48
Speaker
Most photogrammetric programs will have problems with that. Photoscan will not. You can basically tell it, ignore all the stuff in the background, just take a look at the object, and it can take that information and generate a high quality 3D model. More generally, you don't actually need a fancy camera for doing this kind of work. A standard point and shoot can actually give you reasonably good results. Take as many photos as you possibly can,
00:20:19
Speaker
Take them from different positions. Don't stand in one location and kind of rotate your camera because it needs to have the camera moving relative to the object to be able to sort of triangulate the coordinates on the object and generate a full 3D model. If the lighting is bright enough, you should reduce your aperture to as low a level as you can to increase the total depth of field.
00:20:47
Speaker
because if something is out of focus, it's not going to be able to generate a really good 3D model. And lighting is always tricky, especially if it's a sunny day. If I can, I prefer to shoot in cloudy conditions because you get more uniform lighting that way. If that doesn't work, then I will take the photos into a program like Lightroom and boost up the shadow detail a little bit to sort of bring out the missing details.
00:21:15
Speaker
Yeah, a lot of the software actually automatically downsizes photos to a much lower resolution than you give it. For example, Photoscan, if you run on what they call high resolution, it actually divides the total pixel count by a factor of four from the original one. If you do it on Ultra, it does the original one, but Ultra can literally take you a week to process a full model. So it may give you higher resolution, but it doesn't make a lot of sense because it's just going to be too slow.
00:21:45
Speaker
Other programs have a maximum acceptable pixel size for the photographs of about five to six megapixels. I think the largest that 123D catch takes is six megapixels. So shooting at, you know, 24 or 36 megapixels makes absolutely no sense at all. In other words, you don't need a fancy digital SLR, a plain point and shoot that's not too crappy will get you good results. And I've seen people getting reasonable results with just using cell phone photos or even cell phone videos.
00:22:14
Speaker
Now that's really interesting because that's typically something you know every field crew has is you know one point and click at least and then everybody's got a cell phone or smartphone practically. Yeah in fact the 123D catch has a iPhone and Android apps that allow you to take pictures and upload them to the server and then get back a 3D model. So to get started is that kind of the process you would recommend?
Getting Started with Technology & Web Services
00:22:41
Speaker
experimenting with kind of the cameras you already have out in the field and using some of these techniques, but then using a web service like 123DCatch or some of the other programs that you mentioned? That would be a perfect way to start. In fact, that's probably the way that I started. I didn't get the Photoscan software about a year and a half ago, but starting off with 123DCatch, sort of getting a feeling for
00:23:04
Speaker
what photos will give you good results, how to shoot the photos. You know, that kind of experience will be easily translated to more advanced software as you move into it. Yeah, it seems like a great way to kind of prove the utility of that kind of recording method before spending a lot of either grant money or your client's money going out and doing that. Speaking of which, you mentioned that the model that you showed us was from an excavation in the southwest. Was that a cultural resource management excavation?
00:23:32
Speaker
It's a highway expansion that's going to be taking place in the next few years and this Pueblo is probably the largest feature in effective and was excavated over a good part of last year. And so was that type of modeling something the client requested or was that something that you and maybe some of the other CRM archaeologists had to kind of put in the proposal and justify?
00:23:59
Speaker
The client requested, but the CRM company put money into the budget to do this kind of work on that. And then with the resulting model, do you know where that's going to end up as far as the BLM and other maybe DOT transportation organizations? I don't know of anybody that really has a facility set up to store these types of digital models besides burning it to a DVD and putting it on a paper archival shelf.
00:24:29
Speaker
I'm not 100% sure what's going to happen with it. I'm sure it's going to be presented to the client and they'll get a copy of it. There are
00:24:38
Speaker
web services that allow you to put these kind of models up online.
Sharing and Annotating Models on Sketchfab
00:24:42
Speaker
There's a website called sketchfab.com that lets you upload 3D models to share with other people. You can also set them to be private and only shared by a limited number of people. And you can also annotate them as well too. You can put little notes on them and sort of highlight specific features on those. So this would be a good tool for, you know,
00:25:04
Speaker
public outreach or just for public education in general. You know, we can show this is what we're doing and these are what all the little features on this particular model need.
00:25:17
Speaker
That's it for another episode of the archaeotech podcast. Links to some of the items mentioned on the show are in the show notes for this podcast, which can be found at www.archaeologypodcastnetwork.com forward slash archaeotech. If you like the show and want to comment, please do. You can leave comments about this or any other episode on the website or on the iTunes page for this episode.
00:25:37
Speaker
You can also email us at chrisatarchaeologypodcastnetwork.com or use the contact form on the podcast webpage. If you'd like us to answer a question on a future episode, email us. Use the contact form on the website or tweet your questions with the hashtag archaeotech or tag at art pod net in your tweet. Please share the link to this show wherever you saw it. If you'd like to subscribe to this podcast, you can do so on iTunes or on Stitcher Radio. You can also type the name of the podcast into your favorite podcasting app and subscribe that way.
00:26:04
Speaker
Don't forget to go over to iTunes and leave a review of the show. It helps us get noticed so more people can find our podcast and benefit from the content. Also, send us show suggestions and interview suggestions. We want this to be a resource for field technicians everywhere and we want to know what you want to know about.
00:26:26
Speaker
This has been a presentation of the Archaeology Podcast Network. Visit us on the web for show notes and other podcasts at www.archaeologypodcastnetwork.com Contact us at chrisatarchaeologypodcastnetwork.com
00:26:45
Speaker
Thanks again for listening to this episode and for supporting the Archaeology Podcast Network. If you want these shows to keep going, consider becoming a member for just $7.99 US dollars a month. That's cheaper than a venti quad eggnog latte. Go to arcpotnet.com slash members for more info.