ALSAccuracyLive

Hi, I'm going to talk about ALS scanning at M

and G in particularly our high density scanning and what

we're doing as far as accuracy in measuring how accurate

we can get with this new technology.

Um So just as an overview, uh against the four

VQ 4 80 II scanner, which is one of the

newer ones from regal.

Um mhm go into.

So we mounted into assessing the plane.

So it's a very expensive scanner which costs a lot

more than the plane.

So it's an expensive scanner in a relatively cheap plane,

which makes it very efficient, efficient way of getting a

lot of points.

Very economic.

We use a high resolution camera with it, 100 megapixel

hassle. Glad that uses very good high density photos.

Um So let's look first at the density we get

so a few points first.

If we compare it to mls mls is much more

dense. Uh you will get 700 m uh per square

meter in the middle lane.

This is for an older scanner compared to you go

one lane across and you only get 377 points.

So it drops off quite a lot.

Then you go Another even 10 m away Over on

the other side and you're down 22 points.

So it drops off very fast with the mobile, it's

much more dense, but the density because of the height

of the scanner is much much less even.

So the beauty first up of the ALS is where

55 points per meter squared in the center.

That's for one pass, there were only 45 on the

edge. So it drops off slightly but not as much

and then you can increase the density with more pastors

and other things.

So that's intensity.

Um, now the first thing we look at is the

noise of the scanner, The precision, so to speak.

So the density is very good.

What we do is we take a circle of a

plane in the scan and we analyzed the data in

that little circle and you say, uh just from an

older scanners, you can see that the pattern you get

is similar to Mls but the first thing we know

is the accuracy or the precision isn't as good in

that it's plus or minus 15 mil fitting to the

plane of that circle where the mls, Australian red hair

is plus or minus only a few millimeters.

So clearly there's a degradation and that makes sense due

to the, The drop off you have in the coming

from 1000 ft away rather than coming from, you know,

10 m.

So for a high density scanner, this sort of one

m circle looks like the number of points going across

it. And that's the thing applying to that little piece

of road.

That's the noise.

So you expect the plane's gonna be only the road

is going to be flat within a couple of millimeters.

But we're still getting plus minus 15 million points compare

that. But the beautiful thing is because it is a

plane, you can average those points and just a moving

average, you can bring the actually down quite easily.

So precision can be improved when you average it over

a flat plane.

So that's one thing to know.

Um Here's another one, the same data.

If you look at across the road, you can see

it graphically that the mls in green is very flat,

where the average of the red is about the same.

If you took, an average of both of them are

very similar within a couple of millimeters, but the spread

of the red is much bigger.

So uh again, you can see it easily in the

mLS data versus the ALS dollar.

So how does that work when we look at accuracy?

If the noise from the mls is to mills versus

a L S seven mil.

But the accuracy total of the system is not nothing

to do with the precision.

It's to do with the GPS and multi path and

other things.

That's a 15 male for uh uncorrected or unpinned data.

And so the total accuracy of the system through propagation

of variances is not much different.

6, 16 Mil vs 50 male.

Even though the as we've said, the ALS is less

accurate or less precise.

Um Now if we the way that they correct mLS

data is by tapping into control every 200 m even.

And you can bring the accuracy down of the mls

25 million height.

Uh Once email by doing that.

So when you propagate the variances of the five mil

accuracy with the different decisions, the mls will still become

more accurate, but not as much so in a three

sigma case, uh the mls is sub two centimeters where

the sub three centimeters, so that's directly what we get.

But also if we take a disc on the road

and average them out first, we get the same precision

as the mLS.

So naturally as you expect, we get the same accuracy.

We can measure the accuracy of a disk to the

same level as a mls accuracy.

So that's just so theoretical.

Uh And so the point is, can we prove it?

That's where we looked at with some testing.

So the first test is in what we in the

freeway here in Perth, it's 10 km of a row

of the main freeway.

Any north, we do six passes of mls and we

do three passes of ALs, you can see the it's

a nice straight road with the mls we take, because

we're running down the road from the trajectory, we take

a point every meter and we can get uh use

that as a test point to compare between each survey

only for height.

Now, here's the results from mls, you can see the

six passes.

Uh they go out a little bit.

This one obviously got some multi path error in it,

but most of them stick together pretty well near the

end, we've got some spread as well up to three

or four centimeters and that's typical of what we've always

seen with less.

And that's part of the job of picking to control

or averaging, bring your Mls points out together.

Now the ALS is very similar.

It's just you can see the precision from coming from

300 m in the air is less precise, but you

get the same accuracy.

So it shows when the data than what we're saying

that the accuracy is about the same, but the precision

is bigger with the girls.

Uh here's to on the same graph.

Uh and you can see the average around about the

same point and that's just showing them that averaged once

you pin them all together with a L.

S, pin them together with L.

S. M.

L. S.

Uh You get similar results in the accuracy.

So statistically the numbers show that for each of the

less passes when compared to the mls mean, uh we're

getting About five mil in stand deviation over those points.

So there's thousands of points when we get through the

sand aviation of five mil.

And this is for an average rectangle of 27 points.

So it's half a meter wide over a meter long

for each one of these points.

Uh So it sort of proves what we were saying

in theory.

Uh This little test down the freeway, we think we

can get five more.

Uh, We can measure an area, a flat patch on

the road and get five million accuracy, one sigma.

Um, And if you take any one point we can

get uh, plus and minus or better than 30 mil.

Oh, that's racing.

That's the Mitchell freeway test.

Another test we did was Kings Park.

This is like a big area.

It's one of the main parks in Perth.

Um, yeah, if you come to birth, it's one of

the nicest places to say.

Uh, It's mostly trees with a few roads going around

it. So, uh, in those, we've got 11 control points

we want to pin to, we have 23 verification points

are independent.

You can see them there roughly.

Uh, each control point, just the way it happens is

a quite point.

So it's four separate points on the road, but they're

so close, we treat them as one.

So here are the graph of all the residuals at

each point after being pinned at those 11 points.

Uh, it should be noted that the MLS is compromised

all the trees.

So the analyst never came up that well, but the

control points well done traditionally.

Mm Yeah.

And yeah, so the, the answer was, we got five

mil, uh, this way too, so that it did prove

the results of the other test.

Um, so in general, uh, That's the results.

We can get five million one singer or, Or some

30 mil for an individual point.

Yeah. Uh some of the problems obviously we've gone through

the noises biggest, that's why one point is less accurate.

Um the circle, the beans themselves because of coming from

such a height.

Uh the footprint of the beam when hits the ground

is bigger.

It's about 12 centimeters wide when it hits the, the

bitch man.

So picking up curbs and small features, it's not going

to be ever as good as mLS coming from 10

years away.

The other big draw card or disappointment, it's not very

good on the bridges.

So that's the main difference.

Um another thing to look at is horizontal control.

If you want to pin a point cloud together horizontally.

Uh it's not quite as easy to do.

Even with this level of density, you can certainly see

features but to actually match them and be better if

the computer does by drawing off lots of lines between,

let's say I am an mLS point cloud and the

china always pointed out so the computer can do it.

That's more work to be done there.

Finally, another years we've found for this stuff is if

you want to do is to a large area, you

need a lot of verification points.

One we looked at doing is with VLS, which is

just dynamic TLS, we have a L S scanner on

top of the vehicle and as we stop, we do

a scan of the area and you can pin it

using gps over that five minutes of the scan.

And so you get and you're getting um RTK or

slightly better than RTK accuracies with the five minutes can.

Yes but it's enough to verify the sub three cm

single points of an al.

Scan. Uh Little thing we found is because the this

is the A.

L. S.

M. L.

S. And the BLS points together and you can see

they all track together pretty well.

And that's that's because we're the other areas would be

g opioids and control errors.

Mhm. Ah So there's a workflow to use that so

we can do it fairly big area with a less

fairly um Some three cm fairly efficiently.

Mhm. So I'll leave it there.

Um That's what we've learned today.

Uh Ls in summary the LS is more accurate than

we thought it was and when used correctly it can

be done Sub three cm or even sub two cm

and used for checks if you're taking discs on the

road. Uh Thank you.

Hope you got to make from it

ALSAccuracyLive

Chapters

Audio TranscriptAuto-generated