Point Cloud Extraction for Infrastructure Projects

KEVIN GROVER: All right. Sorry for making you sit here for so long, we figured we might as well wait until pretty much 8:00 so we're just a little bit-- Good morning. Welcome to the first presentation to officially kick things off, so bright group eyes this morning, which is good. Better than Thursday morning. So thanks for coming over.

My name is Kevin Grover. I'm region of technology lead with Stantec in Edmonton, Alberta, Canada. My background is a land surveyor. Been involved with point cloud collection for the last eight or nine years. We're into drones and other types of terrestrial-- kind of got into mobile collection, aerial collection, all that kind of stuff. So we do pretty much the whole gamut of things. The presentation today is going to be on the extraction side of things, specifically for point clouds, and tailored to infrastructure projects.

So co-presenter is Ramesh with Autodesk. He's the principal research engineer for Autodesk. Do you want to do a bit of a bio? No. OK. If anybody's been using the InfraWorks products with the point cloud capabilities for the extraction functionality, he's the brains behind it. So we're going to be going through some of that. We're going to break it up into two sections. Overview from my start of the beginning of just some best practices, how we can leverage the data, how we can use some different tools we have today. And then we're going to get into some of the live side of things to see the InfraWorks in action.

Maybe I'll just get a feel for the audience here. I'm assuming everybody knows point clouds by now if you've been to AU. Does anybody collect their own point clouds? There SE are all end users of point cloud data? And extraction? OK. How many have actually used the extraction tools within InfraWorks to this point? So a few. OK. Perfect.

So this class is going to be talking, specifically, how to more effectively use point cloud data. If you've used point cloud data you realize it's a real challenge to deal with. There are big data, especially for infrastructure projects, for large corridors, that kind of thing. It's a lot of information to manage and handle on a project, and utilize effectively. So this is what we're going to focus on is, how to effectively utilize that data.

So my handout that we kind of did before-- We're not to go into all the detail of all the nuances of data exchange between Civil and everything. So we're going to focus more on just how to actually use the point cloud-- Some best practices for collection, and the types of technology actually used for some of the data collection itself. And the benefits of how those will roll into more of the automated extraction. And then will get into the end part of the modeling infrastructure assets on automatic ground classification, and the InfraWorks tools.

So we'll start with just how to utilize point clouds for visualization. So if you've used recap and all the styling, I'm not going to go into normal view and intensities, all that kind of thing. I'm hoping we're past that point. And again, I'm not going to bore you with the technology itself. You've probably seen so many ways to collect the data. I'm just specifically going to be talking today about terrestrial laser scanning, which is a big part of our business in our company. We've got 18 scanners in our firm, and primarily we do the majority of our own data collection.

Also going to jump into some mobile data collection. I'm not sure how many people actually use a lot of mobile data, but it's kind of the Cadillac of the data collection technology for point clouds, especially for urban environments, corridors, that kind of thing. So they can collect it at faster driving speeds. Of course, there are hand-held data collection. There's aerial LiDAR and of course, drones-- which, I run our drone program in our company, too. So I'm not going to be talking about drones. There's enough presentations on that, as well. And they're not typically lighter anyways, so we're focusing more on just terrestrial mobiles. And they're typically the two predominant technologies-- more so for corridor and infrastructure-type work.

So just a bit of comparisons-- I'm not sure if you're familiar with the two sides of it. But just kind of going over a couple. So terrestrial, of course, it's less expensive. People don't think it's cheap, but it actually is. I guess the $50,000 system as opposed to-- the picture that you saw there was about a $1 and a 1/2 million system. Just very different data collection methodologies because the mobile system is actually moving when it collects the data. It's got a lot more sensors involved with it, so of course the cost gets more expensive.

If you're just kind of an end user of data products, you'll also find that there's a lot more data vendors out there that are actually doing terrestrial data collection, as opposed to mobile. Purely just based on cost. There's a lot of other nuances along the way for the data processing methodology that's a little easier from the terrestrial side. But for the mobile side, there's more that are popping up. But not all of them are built the same either.

Accuracy-- this is kind of one that's subject to so many different things. So terrestrial is typically more higher accuracy because we can actually QAed a lot better, integrate it with better survey control, and you can just get much better information, in general. But mobile has come a long ways, and depending on the system-- That one that you saw there, as well, it can collect at highway speeds and you can get down to a couple of centimeters accuracy, even driving at 60 miles an hour. But that also is dependent on the methodology, and processing, and the vendor and all that kind of thing too. So there's a lot of nuances with it but they're both very good for feature extraction for ground terrain, as well as leading infrastructure information.

I'll get in to this a little bit-- Terrestrial has a lot of line of sight issues, especially when you're working on, say, an urban roadway. As you're scanning, there's constantly cars driving by. There's other features that are blocking your way, whether there are power poles, trees, and where you're set up-- That's typically why it's kind of challenging more on a corridor type thing to use terrestrial scanning, because you're bouncing around and trying to collect from different locations. And you're getting a lot of gaps in your data.

Problem with some of the gaps-- which I'll talk about as well-- is it makes some of the automation a lot more challenging. Manual extraction can be very easily done with it, because you're always going to kind of work across some of these voids. But for the automation, and some of the algorithms, they need to have a very linear, seamless data set.

So that's kind of where mobile does have its advantages, because it can actually collect at faster speeds. You can work in, and around, traffic. It doesn't look like a bunch of noise on the road. It can actually follow the speed of the road. As long you're not sitting next to a car when you're collecting data, then you'll actually be able to see the whole road corridor.

Some of the big ones to us is just safety issues in general. So terrestrial scanning-- of course, it has its safety nuances, or safety benefits, well over traditional ground survey methodology because we're not physically entering the road. But you're still physically setting up on the side of a corridor, the side of a roadway. So there are some challenges with it, as well.

The benefit, again, in mobile is being able to do that at highway speeds as you're driving. So you kind of take some of the safety aspect out of it with the mobile collection. Longer time to collect for terrestrial because you're physically sitting static and moving from point to point. Mobile you can collect miles of corridor in a matter of hours. It's ridiculous how fast those systems are collecting data now.

Again, more about the nuances of the two. So from the terrestrial side, everybody's seen the point cloud. We tend to primarily work in just intensity views. I don't know, RGB is-- it looks nice but it just costs too much money, and time, and effort to actually collect it. Especially for some of these projects.

So this is just a single scan location. So you'll see this is when we get some of the void things. Our scan location is here. You can't see across the road. We can see the curb on this side but not on the other side. So again, from the line of sight challenge thing is, you're set up at a certain height and you can only see so many features from one given point.

You add a second scan into it, and of course, it starts getting a little bit more complete. But it's still done on the one side of the road. If we had another scan in, jump the other side, of course, notice this is when you start getting more completeness of your data set. So every time that we start doing this it does add time in general, because you're doing more set-ups And more strategic locations, to be able to see all of the areas you're trying to extract.

So this is kind of one of the challenges again, with terrestrial. When you start putting all of the scans together, you'll still see that there's always going to be little voids in your data set. You know, honestly, it could just be because there's a parked car. It could be just because you can't physically see another location. So all these things do come into play from the automation side that some of the gaps in the data are going to be a challenge to do some of the extraction in. But again, you get you can still get a very robust data set with terrestrial.

So this is some of the gaps thing we're looking at. And this is where it is challenging from the automation too. So we're set up on this side. Of course there's a pole in the way and you get this massive gap of data just on a section of the roadway. So if you zoom in to that-- you can see it-- your dense point cloud of your curb stops here. You've got very little data in between and then it stops on the other side.

So if you're manually extracting features, that actually is pretty easy because you can just work your way across here. Pick the point, pick the point. But from the automation and the algorithms, it'll basically just stop at those locations. It doesn't know how to basically extract what's not there. I know they're trying to come up with that, to do some of the occluded areas, but that's really challenging and more of an infrastructure environment.

So, of course, in the end the intent is to get one, big, complete data set. It looks really ugly because this is just part of the game when you're actually working on an infrastructure project, or a roadway project. All that crap you see, is all just car noise. So, again, it takes a while to clean all this messy stuff out, whether it's people or vehicles. So this is where we can do some manual cleanup of this to kick it out. Or we can throw it into some software to actually just work around it. You don't actually physically have to clean up a lot of that noise information if we want to just derive a terrain surface, and that kind of thing too. And that'll be part of the demo part in the end.

So now for the mobile side. We love it. It's something that we don't technically do in our company yet. It's kind of part of the next area we're going to get into. Our challenge in Canada, we got six months of snow out of the year. So it's kind of a paper weight for six months. So in other offices you might be looking at doing it.

But some of the nice things about it-- when I talk about how linear a data set. So this is actually just a single pass on the roadway from a single scan head. So some of the scanners-- the mobile systems-- have two scan heads. They're typically fastened like a crisscross pattern. So they are collecting at two different angles that's actually driving. So the benefit of that is it's actually getting all the information as it's driving down the road, and kind of getting into all the nooks and crannies that are needed.

So this is just a single pass here. So you can see that there's a very linear data set. That that's the scan line. So it's actually driving and the scanner is actually oriented at about a 45 degree angle. So once you add in a second scan-- so now this is the second one in the opposite direction. You'll see now the scan lines are going this way. But you see how just linear the entire data set is. Everything looks very clean because there's very little noise, as long as you're not following a car. It's actually collecting the data as you're driving.

You put those two data together and now that's when you start seeing this real crisscross pattern in your data set. So you get this incredibly dense point cloud of a roadway. And you get all the features that are even outside of it too. And this is down to a couple centimeters accuracy for specific systems. There are some systems that aren't nearly that good. But like I said, the Cadillac one, the RIEGL system, just over a $1 million, $1 and 1/2 million. It's incredibly phenomenal accuracy, overall, for that system.

When you drive to pass in the opposite direction-- So now we've got the two corridors. This is in two passes. So this is just showing the opposite side here, as well. You've got two passes in both sides of the road. This is where you fill in every little nook and cranny along the way. So you get curve on both sides of the road. You get a very seamless ground surface. And there's very little noise even right off the bat. So this is kind of a benefit with the technology, just in general, over terrestrial, because you get a much cleaner data set right off the hop.

If you kind of look at that in a bit of a cross section view, you can just see how clean all the features are. The paint lines show up just perfectly. It's a very nice data set.

So that's kind of just more about the two types of technology. Then we can kind of get into some best practices of how we actually manage and work with the data to augment some of the data collection itself, and the automated extraction. One of the big ones is to unify, or not to unify. I'm assuming that most you know what unification of the point clouds is.

For terrestrial data, we prefer not to unify in a lot of cases. Just because we like having this scan location information. Just for the ease of filtering things, you also get that kind of bubble view of your scan view that looks like a 360 panoramic image. You can do a lot with that type of data. As soon as you unify the data, and basically merge it, and average it all together-- you lose the scan locations.

I know it works a lot better in the InfraWorks side, when you do get a unified data set just because it does simplify. You put less data in so the algorithms don't have to work so hard to basically filter out even more of the data it doesn't actually need. It is a smaller data set, so of course, all your Autodesk products will work a lot better with a unified data set. But it depends what you're going to do. For the building side-- for whenever we working in Revit, et cetera-- we never will unify just because we like having the scan locations. But in this case, you've got your two sides.

Tiling a data set for corridor work. This is definitely a no-brainer. If you're working on a long linear corridor that's-- whatever, 50 miles long-- you don't want to have one giant point cloud that's 50 miles long. Ideally, you want to break that up into chunks.

There is some regioning functionality within Recap. It's OK, but you might also want to simplify it even beforehand, and tile it even outside of the products. And just bring them in chunks, whether it's a mile long, or a half-mile long. And then you just load data as it's needed. Because the benefit, again, in the InfraWorks automatic extraction, you can load as many files as you want and just process individual ones. So you don't have to have it all loaded into one project.

Of course, the last one-- we struggle with this in our company, just for data size and data, in general. We've got a massive amount of portable hard drives kicking around all of our offices. Our IT guys despise us.

We just did a count the other day and I think we've got 42 terabytes of data from the last year and a half. And it's just sitting on hard drives or computers in our office. And that's not even in the mobile collection side. Once you get into that side, you could technically collect about a terabyte in a single project, just from a couple of hours of driving between point clouds imagery.

So big data is a big challenge. We can't run things on a network. We typically do everything, locally, off of a hard drive.

We're pressuring Autodesk to come up with some better ways for that, because of course, that's kind of some of the challenges. But that's also where we get into-- only bring in the data that's actually necessary. Only work with the data that's actually needed. And then through a lot of the filtering process-- the information as well. We'll filter it down to the point where it has the same visual fidelity. You can get the same data out of it, but it might be a tenth of the file size because you're just filtering a lot of information out.

I'm kind of getting into just some of the methodology of how to generate information out of this. If you've done some extraction you know how painful it can be to do a surface extraction, or even just breakline extraction. If you don't have the right tools to do it, manual modeling is a real nuisance. But it's, by far, one of the easiest ways of doing things because the tools are kind of built into Autodesk products right now.

So what I mean by manual-- this is just in AutoCAD-- just a basic view. A little bit of an oblique view of point cloud. And your basically going through and just tracing features, or adding points. It's like gouging your eyes out doing this type of manual modeling. I don't know if you've done it. It's not fun. It's not pretty.

You're trying to extract all these features-- breaklines for bottom of curb, top of curb. In this case, we're picking off vertical features or manholes. You just need good software to allow you to do this. But anybody can do this. There's snapping tools on Autodesk products. You can kind of manual model some stuff in InfraWorks. It's not the most glamorous way of doing it. So that's kind of why we're always looking at pressuring software companies to do things a little bit better.

We can kind of get into the next level of things and how we do a lot of extraction and other ways of doing it. It's more of an assisted manual workflow. And this is actually a lot more of our preference. The automation side is going to be a lot better in the future, and we trust it. Or we'll trust it a lot more as the tools in Q8 grow.

But this is really nice because it is some manual effort, but it's using some more intelligence. So for like a linear corridor-- what we typically do in a station like that-- we're working in more of a cross-section of view. I'm not sure if anybody's worked in a view like this, where this is actually in Cyclone product. It's just the screenshot that we had at the time, with the codes that were showing stuff.

Basically, if you've got a linear corridor you create an alignment file. You create your cross-sections every 10 meters along the corridor. And you're just cutting a slice of the point cloud, every section along the corridor. So you get this really nice cross section of the roadway. And there's a lot of intelligence software now, where we can basically take these data sets and you can manually start coding or extracting 3-D features, and picking off all the features in a cross-section of view.

As you hit your next station 10 meters away, it leaves those point locations there that you actually had from the previous time, and you're just tweaking those locations. It's a really fast and efficient way of doing it. It's not as monotonous as you physically tracing and make it to your snapping to an actual node. This more of an assisted mode, for us, is really ideal. Because we can actually Q8 the data as we're going.

Manual, of course, you're doing it as you're doing it, as well. But in this case, every section you go along you'll know that all the features you're actually creating-- especially if you've got just a long, straight corridor-- you're just picking up very nice features along the way. You've got almost perfectly clean, three-dimensional polylines, or features along the entire corridor.

Then from the terrain side we've got different options. And I know, ideally, it's the easiest just to throw a whole shitpile of points into InfraWorks or Civil 3D and let it create a surface, and then you've got something. That's not the most efficient way. Of course, if you work with Civil 3D-- or even InfraWorks or any product-- the more data you throw into it, and the more data you have in a surface, the harder it is to use.

Civil 3D-- it's gotten a lot better at handling surface data. But you definitely want to have a lighter weight surface, especially if you're going to look at things in three dimensions, or start working with it after the fact. We've got a couple of different ways you can do this. Again, the top one is typically more our preference. Or the workflows that we have now, is to create a surface from all your 3-D features.

So if you're extracting just all your 3-D polylines, and breaklines along a corridor, you're actually just creating your surface out of those three dimensional features. So it creates a very nice, seamless surface. It's not very dense because you're only doing it to the density of whenever your cross-sections are. It's very lightweight, and it's clean, but it's very time consuming, because you are physically just creating these features you need along the way.

Second part is to create a surface from a point cloud. Of course, this is kind of the holy grail of it, where you've either got to manually clean it enough to get all the data out of it-- There's some tools in Civil that are in the handout I talked about. Some of the ground extraction functionality in Civil-- it's OK-- but you're still left with all these kind of spikes and peaks within the surface. And it is based on only what you can see too. So it is a lot of editing that you have to come up with as well.

The last one-- You can do it just off a grid of points. So we will typically do, is from our aerial collection, or if an area is completely flat-- I want a 5 by 5 meter grid, and that's kind of the surface. So it's just going to randomly pick points every five meters, and you can generate a surface out of that. Again, very lightweight. You miss a lot of the detail in your surface, especially if you've got curb lines, and maybe a little narrow drainage courses, that they won't actually be picked up that way. So there's a lot of different ways you can create a surface and they all kind of react a little bit differently in the software.

So now we're going to get into more of the automated one. And the crux of this presentation is how we can start leveraging technology and computer vision, which is-- and I'll throw this in, Ramesh will like this-- is his magical computer vision. I don't understand all the computer vision side of things. They've done a lot of very impressive algorithms to start working with raw data sets and kicking usable products out of it.

I'm going to hand this over to Ramesh to kind of jump in for part of this. This will kind of get in the next two sections of how we're using the data-- not so much putting in InfraWorks in Civil 3D, but just purely an InfraWorks we're going to show you all the extraction tools, and point cloud functionality. And then how we can actually use it to automatically model infrastructure, right within InfraWorks.

RAMESH SRIDHARAN: Thanks Kevin. Let me see if I can pull up-- [INAUDIBLE]

So it's always interesting working with Kevin. So you start a project as a simple project-- using this data he said, there's the intersection project data we have, let's work on it. And then I open the data, there's a lot of cars on the street, and there are some glass buildings and stuff. I'm like, something's wrong with the data. And he says, OK, let me give you some other simple data. And that was this. A lot of cars again, in the noisy environment. It's always interesting.

The duplicate doesn't work. [INAUDIBLE] works. I'll try to work it out.

OK. So like Kevin said. the data has lot of-- I don't want to call it [INAUDIBLE], I mean they are features for those who are interested in it. But it has a lot of points for the regular project. If you want to create a surface, if you want to extract some features out of it, then all these things come in the way.

Even if you do it manually, it's a very tedious process to go through it and also, there's an easy chance you might end up missing something when you do the manual process. And that's not specific for this particular data set. At any point [INAUDIBLE] data set is the same thing. So the more you work with it, there's the possibility of making mistakes. It's a tedious process on that part.

So when we started adding tools [INAUDIBLE] in Autodesk product we were thinking how to handle this particular part better. Let me actually show you my part one, then I'll come back to this data to give you-- It's kind of hard to maneuver. This way.

So when we started adding the tools in Autodesk product we were thinking how to handle this particular situation, because it's-- how do I say this? It's not that hard to create a cross-section based tool and go extract the vertices, just like everyone else is doing. But we thought can we do a little bit better than that? So that it's actually a much better workflow for the modeling, as well as the design perspective, and provide a completely end-to-end solution for a large data set. Not just the intersection, not just the road-- liberal, large data set.

So that's where we came up with the concept of the end-to-end solution to handle this. This part I'm going to skip because Kevin already talked about it. So that's where they came up with the concept called Points to Models. So completely putting myself in a customer perspective, you are sitting on top of a lot of point cloud data. And yes, you go through a manual way of extracting some lines from it, and you're doing it, but point cloud has much more to offer. Just because we don't have the tool doesn't mean that you cannot use it in thing.

So having the capability like converting a bunch of points to corresponding models in a reality world, makes the worth of point cloud much, much more. And you can do things which you never imagined you can do with it. That's when we came up with this concept. So how to do that?

When I am talking about automated tools, then all the nicest-- in the middle of the road, and side of the road-- all of this comes in the [INAUDIBLE], it's very hard to do something. So the first basic thing we have to do is filter out the data. And it has to be automatic. We cannot expect users to come up with a paint brush kind of a thing you go over and filter it.

So we came up with a concept called extracting the data derivatives. Extract something that's a sense in the point cloud so that, number one, we have the information content. And number two, we can use that to filter out the stuff so you get to see things, or do things with it what you want, instead of loading up with gigabytes of data sets. Once you have that-- and the rest is actually very simple, it actually paves the way for everything else.

My really little favorite is-- Which one is a laser pointer here?

KEVIN GROVER: Yep. Just the red button.

RAMESH SRIDHARAN: Oh, this one. So my favorite is the vertical clearances, and on the route modeling, and things like that. So route modeling, for example-- I was working with the Oregon DOT. If you want to go from point A to point B, right now the big trucks, the [INAUDIBLE] trucks have to get the permission from the DOT. And the only way they're doing it is, they have a picture-- like a Google image kind of a picture-- and they have the bridge and there's a sign-- 16 feet or something.

The person who authorizes the route just looks at the picture and says, well don't go this way, go this way. But it's a super elevation. You can actually try to go in one lane and you can get there if you want to, but that's not the way the system works right now. So we are losing a lot of efficiency, even though Oregon DOT has a lot of mobile LiDAR data. They are sitting on top of the data. Still, the ability to get the route modeling from point A to point B with your 3D data sets is not viable right now. So that's exactly what we are trying to change with our automated process.

So what is it that we do? So basically, this is a typical mobile LiDAR data. This is not Kevin's though. It was a different data. So we bring in this data. What we want to do is to remove the nicest first. Generate something bare earth. And I kind of like the concept of bare earth. I know this bare earth is not the exact thing you will use in design, but this is the bare earth people started collecting point cloud data, 15 years back, 10 years back.

So now when we start using it, we always go to the line worked interpolate between them. In my mind, it cannot be wide, the purpose. I understand the purpose. Don't get me wrong. But there are much more in the point cloud data. Like a bird bath in the middle of the road, you might miss when you do a cross section based approach or something. Maybe you don't want that for your project, but it happens. It's there in the data, it's useful for you. So having this bare earth, removing all of the noises, and provides you a clear solution. You can use it. If you want, you can use it. It's much, much better. And that's just the tip of the iceberg. That's number one.

Number two is, we started with the vertical asset. Since we started working on the InfraWorks-- InfraWorks, when you see it, is a lot of 3D models. And it's kind of easy to do that part first-- to get the attention of the customer, then we go to the linear stuff and things like that. So that's the approach we took.

So the vertical features-- this is it behind the screen-- just to let you know-- but you can see that software goes ahead and tries to group each and every feature so that we can use some machine learning, or some kind of approach to ease the user part. My philosophy on this part is, user comes across as the expert. You tell the software what you want to do, but I don't want you guys to do every single thing. I want software to do it. You paid big machines to do that, lets make use of it, obviously. And that's the second part. So vertical features.

And the third one is the linear features. And we are working on it. And like Kevin said, there are some snapping tools and everything, but again, that's a lot of data. You are to go through it in here. The software highlights some of-- the pinstripes are very easy to see. I think I used this slide before. So there's the curbs here, top and bottom of the curbs. You can see it.

The software classifies the point cloud data. High density point cloud data here, so that when you're doing-- even the worst case scenario-- I'm not saying you have to do it, manually. I know that's the only tool we have right now. But even if you are doing it manually, this simplifies your job. You'll just highlight what you want. You can snap through and move on instead of messing with large point cloud data. That's the key here.

So once we have these three-- that's why I call it a derivatives. I know the data is a raw data here, but the concept is applicable for other point clouds also. Infrastructure point clouds, at least. It could be a rails or it could be a power line you are working on, but the concept is exactly the same. Same workflow works for all different types of projects. Workflow works for all different types of point cloud also.

Maybe it's a mobile. In this case it's a mobile. Kevin's project we're going to show is a static there's still LiDAR data. We have some customers who work with the UAV data. UAV is a little bit odd with the point cloud but it does. It does the job much better than-- if you're going to build with the UAC data by itself, it's a very tough thing to do.

So let me just show that once. The large data is the large data. As much as we try to ignore that when it comes to point cloud data, it is large. And you are sitting on-- like exactly Kevin said-- you're sitting on top of hard drives. So we are trying to see, once they do the information content it leads the way there. Why can't they just store the information, and for lack of a better word, pass everything else? I mean, you have the raw data if you want to, it's always there.

So can we do something like that so it makes it better for the processing perspective? It's not just for the-- Of course, it makes software job much easier. You need to show only a few points and maneuver a few points. In my mind, it makes user's job easier also. You're looking at what you want to see and get your job done, instead of looking at millions and millions of points. I don't see the purpose of it, in a way. You know what I'm saying.

So that's when he came up with a lightweight point cloud data. So the one on the left here, is an original one. The one on the right is the lightweight one. I kind of like this slide. I use this slide before. But this kind of conveys the point of what I'm trying to say. That both carries the same information. Both has all the points. In my mind, this actually looks a little bit better than the other one.

But the cool thing is that the one is just a 1/10 of the size. And I'm not saying you're going to get 1/10 on every single data. It depends upon [INAUDIBLE] resolution, and all those things. But this does break it down significantly. Even though I couldn't convince Kevin yet that he can use this. But it does the job. It does the job for you. It makes it much easier. You can share the models very easily.

In a bigger picture-- In my mind, this is how it works. Let's say you have a project-- let's take a simple project of 50 gigabytes of data, for example. And you run through a processing software. It processes it. Let's say it brings it down to-- let's say 5 gigabyte, I'm not even going to exaggerate-- let 5 gigabytes for an example. Then how easy for you to share your model, share your project, with your customers, or with your colleagues, across the country, or across the world, with just a 2GB or 3GB of data on the cloud, instead of 50 GB of data on the cloud.

It makes your project much more easier to handle. That's my opinion. And we are making efforts on that to get better and better. As a matter of fact, some of the slides we are using here on the project was-- it was a little bit of pre-discussion software. The new version makes it much better. I know only a few hands came up when using InfraWorks. You guys should try it out. It makes your job much easier with the point cloud date. Whether you collect it or consume it, it doesn't matter.

So the point cloud terrain generation-- So like I said, you can use different types of data, it doesn't matter. It creates in Infraworks. It creates a wireframe brazier kind of a looking model for you to put your models on. Same thing translates much better in Civil 3D or AutoCAD, as well.

This slide, actually, I had this slide also yesterday-- I didn't like this data. Because it kind of conveys the big slope areas, how to do the terrain extraction. Like for example, this is the one on the previous that are data, and once you process it, it removes a lot of niceness on the data. And there's mainly some of the vegetation points around here. Then, when it overlaid terrain on top of this, it looks like this. And then here's the contours.

And I'm sure there are a few points here and there-- there could be some trees, or here, somewhere-- But this is a drone data. And you can actually use this and create a better surface. Use for the model-- I mean, it's a Civil 3D. I can use the word design, but drone data on its end has completely different things, like Kevin mentioned. I'm not going to go there. But the point is, you can use the point cloud data. You can filter it out and use it with very little effort, is what makes this tool much, much better.

Well let me just play this across-- Whoop, not that. How do I play this?

AUDIENCE: [INAUDIBLE]

RAMESH SRIDHARAN: Yeah, but my mouse--

AUDIENCE: [INAUDIBLE]

RAMESH SRIDHARAN: But I don't see my mouse pointer. Yeah, but-- Oh there's the mouse. That's why.

So this just shows the crux of what we have in the product right now. And you get the idea where we are going to. Point cloud, like I said, reiterating lots of stuff. This happens automatically. No manual work involved in here. It creates a terrain like this. Imagine having a terrain like this in your project, enforce it with the proper breaklines, makes it much, much more useful-- your data.

And this is the model-- the one thing I really like is that when it comes to reality capture, creating a models with real information-- signs or the street lights-- wherever you say there is a street light, there is a street light in the real world. And I know the modeling doesn't translate into design. In my mind, design and modeling are close.

Modeling is just a way of representing in a nice 3D model where the points-- the view just converted to X-Y-Z. That's the design point, that's the design line. Both are pretty much the same, just the interpretation of how you look at the data. So modeling-- once I have this in a model, imagine I can export this as-- I don't have a line here, obviously. If I have a line that's a poly vertices, or the features as a points, those are [INAUDIBLE] point. Those are the extractable points. Those are the feature lines you can create your design with.

So in just the interpretation, but the modeling way of working with the point cloud data makes more, more productivity from the customer perspective. The usage perspective. You're looking at-- well, I don't want to call it eye candy, but it does makes it much easier to look at.

How we do it? I'll just show you a real quick video on how exactly this works. This is a complete workflow. At least for what I'm showing right now. [INAUDIBLE] in point cloud that comes in, it goes into ReCap, obviously. That's a set, you can prep them. There are some [INAUDIBLE] light out, you can register them. Or sometimes I used to clip out some of the areas I don't want, because mobile light out, or something, collects too much information than I want to.

So here this is about-- this is a West Virginia ridge. There were two kilometers of data. You can go ahead-- in this case I'm clipping it-- so my point is, you can prep the data. Here's your point cloud file. In InfraWorks you bring it here, you import it, all the corners from the set. This part goes a little fast, but bare earth, I'll show you later. So you click this thing, point at the terrain. It creates this [INAUDIBLE]. Same data, there's no mock-up here, just a recording of actually what the software does.

And when you go up in the point cloud data, this is what it actually did to your data. You can have the linear features classified, the vertical features, everything is right there. And this is actually a lightweight, too. You don't see it, but it's actually lightweight data.

And modeling part-- the software autobody recognizes many of them. Like the street lights here, street lights there. Once we have the linear stuff we are going to have the orientation, but for now, the points are already there. Software understands what your group of points actually mean and replaces them with the real world model. And you can change the style, change whatever you want-- depending upon your project, your area, and your data set-- but the software does the heavy lifting for you.

And one thing-- I worked with this kind, like I'm sure Kevin worked too-- I worked with this kind of projects before with a mobile LiDAR, especially for asset collections. It is easy to miss. I know any product of that comes and tells you, I can do it 95% accuracy, 96% accuracy-- what are you going to do with the remaining 4%? You have to go through manually to check what it missed. This one has a little bit more false alarms, but it doesn't miss anything. You get everything you want. So you spend 20 seconds or something, but you get the job done.

This, once you have the data, you can bring it down to Civil 3D. We have a couple of ways to bring it in. I'm not going to go through that here, but you can get it done, you can get a surface. In this case I'm showing you the points. You can bring it in here. On top of it, there's all that. [INAUDIBLE] come through it. The other way of bringing the point cloud to create a terrain for you. So this is what I'm talking about. So once we have the data set, how quickly you can use it, and get it to design, is what makes it more powerful.