So stay tuned and see the accuracy of all these systems. Let’S fly Music, two locations Music, so today, we’re doing all that accuracy on those three different lidar drones, we’re testing, the accuracy we’re here in northern california, and we have an actual land surveyor with us doing ground control. Because not only are we doing this video for you, but we’re also doing a boundary survey as well as a topographic survey for a customer. This is their beautiful property here and at the end of the video we’re, going to be doing a complete review of the data compared to those control points set by the surveyor, so we’re flying the three lidar drones, we’re doing the exact same flight path, the same Height above ground level terrain following so it’s going to be a real apples to apples, comparison and we’re, going to see exactly how do these three different lidar sensors work again, it’s all about accuracy. So today we have a licensed, surveyor land survey. Here this is thomas terwilliger professional licensed land, surveyor thomas. What are you doing here today? Well, i’m, excited to be here: i’m, a california licensed land surveyor, and my job is to conduct the control portion of our aerial survey. We have about 20 ground control points throughout the site that we are going to measure with survey grade gps and so that we can tie together our ground control with the lidar data and minimize errors and have a more accurate uh.

Digital terrain model awesome, thomas, so we’re going to go ahead and get out there lay those area. Control thomas, is to come mark them with his gps equipment. Let’S get going great Music. So this right here this is one of those aerial targets we’re using these to actually tie in with the land surveyor all the lidar data into a measurement here on the ground. So we got to put several of these around the job site and go measure them let’s go all right. This is a good spot. All right put that firmly on the ground. Land surveyor is going to come and mark this point so it’s not just about how many ground control targets you have it’s, also about where you place them so right now, i’m. Placing a few more targets here in the opening, but also matt just walked down here by those tree line right there. He placed a target there. So now we have aerial targets next to heavily vegetated area. We have targets in the wide open area, but almost always we’re placing these targets on a level surface. So here we have these three different lidar sensors. Really. What is the difference now? For one thing, price price is a big difference over here we have a 180 thousand dollar lighter sensor. On this side, the dji one don’t tell him. I told you it’s about 17 000, plus all the accessories we’re looking at about 25 000 and the r2a is about 30 000, with all the accessories about 40 000.

. So we have different prices on these systems, but then also we have a difference in accuracy and usability and how they work and where they work so the l1 on the dji m300. Only for the m300, the minivox over here on the m600 well it’s, really heavy. So it’s not going to fly on these other two drones and then the r2a. It can go in any one of these three drones, so that’s the different operability of the systems, and then you have accuracy and really i’m not going to do a spoiler because we’re about to fly and see what that actually is. So why don’t? We just go ahead and get up in the air and start flying and then see what the accuracy of these different systems are. Let’S. Do it Music. First up, we got the l1 let’s fly Music Applause. You got the regal mini box on the m600. Let’S fly. Applause, Music, Applause, Music, all right and now it’s time to fly the r2a on the m300 that’s. Why? Applause, Music Applause? So we just finished flying all three of these lidar sensors. We had the land surveyor go out and capture those aerial targets. You saw me with those checkerboard patterns now what we’re going to do is go back and process. The three data sets from these systems and we’ll. Compare that data to those aerial targets from the surveyor that way, we’re going to have a good baseline to compare the accuracy of the three different systems and, of course, all this data is gon na be on the rock cloud you guys can see it.

You can share it it’s great, so let’s go back to the office. Now take a look at that data and welcome back to the office. Now before we jump in and compare the accuracy of those three lighter data sets. I went ahead and flew a second flight with all three over the alameda naval base, and this is an old retired airstrip, but it’s a very flat, long, hard surface that’s, going to make an ideal situation to compare the accuracy of those three lighter sensors and i Did one more in that? Well, i actually did 190 more in that i captured 190 ground control points there and honestly in the findings of this data, there’s some very interesting things i’ve seen in one of the lighters, in particular with using those 190 gram control points. So make sure you stay tuned for that part of the video, but first let’s go ahead and jump into that hillside data set i’m going to pull up all three lighter sensors in one visualizer we’re, going to look at cross sections of going through the grass and Seeing that relative precision of the fuzz we’ll see how they line up to those ground control points, we’ll, look at the overall aesthetic of the data sets and then we’ll see the accuracy report, as derived by using the rock cloud to generate a surface from those data. So let’s just jump into it right now, so here i am over on the rock cloud i’m going to pull up this hillside folder, and i can see that i have the three layer datasets in here.

The l1 r2a and minivox let’s go ahead and bring them up in the same visualizer. The first up is the dji l1 data set let’s just take a look at how it aesthetically looks at first. So the first thing you’ll notice is man. The colors look beautiful. Very saturated you know very deep, dark shadows as well. As these greens look great, i mean the trees. Look colorized perfectly and you have. The hillside looks really good too a couple things i’m noticing it does. It looks a little bit fuzzy but really good in general. Let’S compare that to the r2a, so here we have the r2a colorized view and you can see it looks a little overexposed, maybe a little uh unsaturated as compared to the l1. Still good. The alignment looks good, you can see, the trees are colored well, and the grass is where the grass should be so it looks like the colorization is lining up pretty spot on and finally let’s go and take a look at that. So right here with the minivox data and you’ll, see right away. You can see these gray striping on here now. This is actually a problem that the camera field of view is not as wide as the field of view as a laser scanner. So there is points that don’t get colorized, and that was a decision i made not to cut that portion off, because what i wanted to do is i wanted to see the overlap of the lidar strips in order to see the actual accuracy of these.

This different, lighter system, so i left that in there, but you can see you know it’s. It looks good, but you do get that big gray banding, which is pretty annoying let’s, go ahead now and look at the ground control points and see how the light of day sets are lined up to those first up is dji one let’s go and take a Look at this ground control point up here and we’re in the rgb view, and we can see it’s a little bit off to the side in the rgb view, but this is not the way you should be looking at this. What you should be doing is switching over to the intensity view, and you can see right there how that points kind of jumped off to the side now. This is because the rgb camera and the laser scanner aren’t perfectly aligned so there’s a mismatch there and that’s that’s pretty normal. So just a rule of thumb, you should always be looking at intensity view when you’re aligning your lidar data to ground control points. Just always do it, so this right here is the control point. You can see it if i zoom out a little bit, i’ll i’ll adjust the intensity, try, making it pop a little bit better. You can clearly see that’s it right there, two white sides and the black sides – i’m gon na – go ahead and change this over to the gps time view and i’m going to kind of get a profile view of that data set right here you can see.

It looks it looks pretty darn fuzzy right there there’s a lot of stray points below the surface zoom back out. So in this view this is the gps time view this is kind of how it was flown in time. You can see this or the red. The orange the yellow, the green – and these are all individual passes of that lidar scanner, so i can do the same thing and switch over to the r2 at the same spot. So here we have the r2a liner dataset at that same control point, and you can see even on this one. The the control target is off here to the right a little bit. But if i switch over to that intensity view, you’ll see that it then pulls over right in line with that control target. It’S gon na zoom out that’s the target right there and that’s the point right there in the middle of it and if i switch over to gps time kind of do this looking down the barrel just kind of getting idea for the fuzz. Definitely definitely a lot less fuzz than that l1, but we’ll do cross sections here in a second. But first let me pull up the the regal minivox at that same location, so here’s the regal minivox, and we can see on this one right away. You see the data set is a lot more sparse, not as many points and you get. This line line kind of banding effect right there to the right that is in the rgb view, that aerial target, and then i can switch again over to the intensity view, and it certainly is more difficult to see.

But you can see if i move this slider. That right here is the aerial target that we are comparing to, and if i switch over to that gps time and look down the barrel of it, you can see once again much more precise data. So now we just looked at the overall look and feel, and the colorization of these individual data sets we kind of zoomed into a control point and just saw how it lined up in the intensity view versus the rgb view and now let’s grab a cross section Across all three data sets and look at the relative precision. The fuzz of the data sets with respect to each other. So, to do that i’m going to use this quick tools and i’m going to look at the compare view which brings up all three data sets in a false color. So now we can see the regal minivox is in red. The dji l1 is in blue and the rock r2a is in yellow i’m going to come up here into the measure. Pane grab the height profile and we’ll make it basically going right next to that control, point and i’m going to extend it out over here. A big chunk here and we’ll turn off those control points. I’Ll make the width 0.5 feet. Oh that’s, not that many feet 0.5 feet there we go. What we’re going to be looking at here is that relative fuzz, that precision of the individual, lidar, sensors and honestly this is not the best location to do it.

So it’s tall grass, so in that tall grass you’re getting measurements from the top of the grass and going all the way down to the bottom. So there’s some natural fuzz in these data sets so that’s, where the airport data set is going to give us a real good information about what the actual precision of these systems are, but for now let’s go ahead and just take. This is a real world example. So you might as well look at that relative fuzz and then, after that, we’ll look at those control reports based on the surface that was generated from the rock cloud. So here we have that profile, tool and right away. We can see the blue is dominating the view of the profile, and that is the dji l1 and then the regals in red, and then the r2as in yellow so over here is where that control point was that we were just looking at. So if i move this cursor along here, that’s basically right over here let’s go ahead and just zoom in on that location. So that’s right where that control point was so now we can see the blue it’s kind of uh. You know this top to bottom right. There in the fuzz, we have the yellow right here, and then we have the red in the middle of the yellow, get some points above and below the red, the yellow. So this is looking at the fuzz of the individual data sets.

This is kind of an indication of the precision of what the data was captured in so let’s go ahead and just take some rough measurements, i’m. Just going to measure points on the high end compared to the low end measure that deviation we’ll, do it for the l1, then the r2 and then the mini box and pull out my phone here and use the calculator let’s look right, yeah, i don’t know what Do you guys think well, it looks like a bottom surface kind of right there this looks average. I know we got points way up here. I guess we can do those we’ll do the extremes. So i got this one. This is 1715.9, so let’s just say: 5.911 subtract out this one Music it’s hard to tell let’s just do this, one that is 1714.101, so let’s do 4.101 it’s about 1.8 feet right there. Next let’s do the same spot. The r2 that is almost three tenths of a foot right there now again. This is all in that grass, so let’s go ahead and look at the regal as well, so the regal’s in red. So we got a point here and the point there. It looks to be a kind of a spread there’s also. This point up here point down here so that’s rated, almost two tenths of a foot. So what we’re measuring right? There is the precision the fuzz, but on this data set, it was all grass.

So really, we want to be looking at a hard flat surface to measure this accurately and that’s, where that airport data set is going to come in. First let’s go ahead and take a look at what was the surface generated from the rock cloud and what was the accuracy able to be achieved from that surface to ground control points so i’m, going to pull up the gcp reports for all three data sets and We’Ll, take a look at that right now. These ground control point reports are based off of the ground, classified lighter data set and then a surface made from that data set. So a digital elevation model and then the contours join from that. So that surface is being compared to the the ground, control points and that’s, where these accuracies are being determined from. So the first step is the dji l1, with the accuracy report right here and if you scroll down it’s reporting a two tenths of a foot vertical rms accuracy, and that is from that surface that was generated from the rock cloud. You can see all the points here. All this data is available online. Next up, let’s, look at the rock r2a. You have the accuracy report and you’re getting about a tenth of a foot vertical accuracy. On that surface and a point: zero, zero, eight foot mean delta elevation and third up but not least, is the regal minivox. So we have .113 just about a tenth of a foot as well on that vertical accuracy.

Okay, now that we’ve taken a look at those three data sets on that grassy hillside let’s, look at the real meat and potatoes, which is this alameda naval base, and the runway is a very flat surface, which makes it just ideal for looking at accuracies, and i Found some very interesting findings let’s jump into that data set right now. So, if you’re following along, we have this folder right here called airport inside of that i have. Four data sets the l1 r2a the r1a, as well as the minivox i’m going to pop over here to the compare view – and here you can see all those ground control points, there’s 190 of these things all over this data set. You can see all the lane markings were captured and then the the lane, the runway number was captured as well as all of these creases and cracks in the construction of the runway i’m, going to go and open up the lighter view now and take a look At this, so here we have all four of those ladder. Data sets superimposed on top of each other and we’re all in intensity view. So this is actually showing us let’s. Just look at one really fast we’ll. Just look at turn them all off. We’Ve got the r2 here and now this is that intensity view, and you can see the lane markings and if we zoom in, we can see exactly where those control points are captured.

We can even zoom in on single one and just see hey. That point is around the corner and it’s right in the middle of that fuzz. This looks good okay. So now i have all four of these ladder. Data sets pulled up on top of each other and i’ve colorized them by which data set they are using. The compare so the dji ones in red, the rock r2as in blue, the r1as in yellow and then the mini boxes in purple before we jump into looking at that precision here on this hard flat surface, i want to point out something i found very interesting. What i’m going to do is i have a large cross section going across, basically the entire runway, that’s kind of cutting through a lot of time that elapsed in the acquisition. So i found something very interesting in the l1 data set so check this out now, as you see on the top i’m going from left to right – and this is basically the direction i flew. So this is over time. If i look right here, we’ll see that that l1 data set well, it seems to be, it seems, to be high. I mean it looks pretty fuzzy, also but it’s higher than the blue and the yellow and the purple that’s down here at the bottom. So naturally, you’d say hey: why don’t you just lower that data set down and make them all sync up? Well, let’s go ahead and look over here.

Let me zoom out come over here and now. If i look over here well, the data sets low now but again, the r2 r1 and the regal minivox are all basically right on top of each other. So what i’m seeing happen here is some time dependent, long term undulations in the accuracy of the data set. So i can do a one measurement of that fuzz at one point and that’s, going to tell you what that fuzz is, but because this is a long and a very flat hard surface. We reveal some very interesting findings that there’s actually long term undulations in the data set as you go over time. So just looking at one point, that would be one instant in time and we’re going to do that right now, but also this finding that it seems that it’s undulating and it doesn’t seem i’ve done a bunch of cross sections it’s not doing it consistently. So there is some sort of floating going above the data sets and below, and this also checks out with all the ground controls. All the all those points as well you’ll see that if you line the data set to the gcps it’ll be you know high and some of them low on some of them, but it’s not consistent, so that’s a very interesting finding – and you know that’s very interesting. So let’s go ahead and just measure some of this fuzz of the four different data sets and just get an idea of that and disregard the fact that you see one above the other one or below the other one, the r2 r1 and the minivox.

If you look everywhere, they’re smack dab on top of each other and they’re also smack dab on top of all the gcp points, and the l1 will just be kind of undulating above and below that surface, so i’m going to go ahead and zoom in right here. So we can see the red is going from here to here, so that difference is 0.485 feet, it’s about a half a foot of fuzz right there and then, if i look at the r2 in blue zero, nine two just under a tenth of a foot, let’s Look at the mini vox in purple, the mini mux got about a half, a tenth of a foot and the r1. The r1 also got just under a tenth of a foot, and we can just see empirically by looking across this data set. The l1 has some very fuzzy sections and then there’s some sections that are not so fuzzy and then a fuzzy again, but that still doesn’t explain that undulation from high and low so that’s very interesting, very interesting, finding now here’s my conclusion. In short, the r2a and the minivox, these are good for surveying. You can get that tenth of a foot accuracy on that surface model and you can measure that fuzz and feel comfortable that the data that you’re getting is going to be accurate and repeatable the l1. Although it’s very inexpensive and affordable, which is awesome and the colorization looks gorgeous there – is that fuzz that comes to the data set and also that fuzz is moving up and down over time.

So this makes me feel that if you went and flew two times or three times or four times the same spot, it wouldn’t necessarily repeat itself because over time you’re getting some undulations now the l1 is very inexpensive. It does capture good data. If you want to see just the general overview or if you want to capture some more less dense contours, so maybe a two foot contour or a 10 foot, contour no problem all day. But when you talk about survey grade systems, you really need that repeatability and that accuracy, and today we just didn’t, see it. I didn’t see it on either. One of these data sets and let that be what it is. So that is basically my findings from the r2a, the l1 and the regal minivox there’s a large difference in price and when you need accuracy and survey grade you’re going to want to use a minivox or this r2a. And if you just want to have the overall view, if you’re doing something that’s, not so stringent on an accuracy, the l1 is a great option. I hope you guys enjoyed the video now all these data sets are available on the rock cloud. You can click on the link below and view them you can download them. You can draw your own conclusions as well and let me know in the comment sections what you think. I i hope you enjoyed the video.