WEBVTT
Kind: captions
Language: en-US

00:00:00.800 --> 00:00:03.040
Hi, everyone.
Today I’m going to show you

00:00:03.040 --> 00:00:07.760
how you can use the laser scanner in a
newer iOS device to easily create useful

00:00:07.760 --> 00:00:12.536
3D models for paleoseismology
and other geoscience applications.

00:00:12.560 --> 00:00:14.560
If you were at the SCEC
meeting two weeks ago,

00:00:14.560 --> 00:00:17.933
then some of these slides
will be familiar.

00:00:18.480 --> 00:00:22.000
Unfortunately, being in the U.K.,
there aren’t any nearby active faults

00:00:22.000 --> 00:00:25.280
to create a demo for you, but that
also means that I live down the street

00:00:25.280 --> 00:00:29.360
from this Egyptian tomb.
Here, I’m using the 3D Scanner app.

00:00:29.360 --> 00:00:33.280
When scanning, it is important to move
slowly to avoid motion blur and make

00:00:33.280 --> 00:00:38.160
sure the geometry is scanned correctly.
This app creates a point cloud using the

00:00:38.160 --> 00:00:43.840
Lidar scanner and simultaneously creates
a texture using pictures from the camera.

00:00:43.840 --> 00:00:46.720
It then puts these together
to produce a photo-textured

00:00:46.720 --> 00:00:50.536
3D model in centimeter-accurate
real-world units.

00:00:50.560 --> 00:00:53.200
Please check out my website
below if you’d like to explore

00:00:53.200 --> 00:00:56.136
this and some other
models yourself.

00:00:56.160 --> 00:00:59.680
The real-world scanning and processing
time here was exactly four minutes,

00:00:59.680 --> 00:01:02.296
and this video
was sped up 4X.

00:01:02.320 --> 00:01:05.896
Let’s compare this process to
our photogrammetry workflow.

00:01:05.920 --> 00:01:10.560
To produce a scaled SfM model, a
control survey must first be established,

00:01:10.560 --> 00:01:14.640
then photos taken, uploaded to a
powerful computer with expensive

00:01:14.640 --> 00:01:19.336
photogrammetry software and processed
before the model can be viewed.

00:01:19.360 --> 00:01:22.320
The fastest I have ever been
able to produce an SfM model

00:01:22.320 --> 00:01:25.520
was about 30 minutes.
In my experience, it frequently

00:01:25.520 --> 00:01:29.656
takes multiple iterations of both
methods to produce a quality model.

00:01:29.680 --> 00:01:34.560
IOS Lidar has a clear advantage
in time, effort, and accessibility.

00:01:34.560 --> 00:01:37.840
As I’ll show you today, the final
image resolution is acceptable

00:01:37.840 --> 00:01:42.419
but a bit less than that from
a high-quality SfM survey.

00:01:43.360 --> 00:01:46.720
This study by Luetzenburg et al.
was published last month

00:01:46.720 --> 00:01:50.960
in Scientific Reports. They
compared scans to measurements

00:01:50.960 --> 00:01:56.296
of real-world objects into an
SfM model of a seaside cliff.

00:01:56.320 --> 00:02:00.720
They find that the accuracy is about plus
or minus a centimeter for small objects,

00:02:00.720 --> 00:02:05.440
like a chair or box, and for bigger
landscape-size objects, like a large fault

00:02:05.440 --> 00:02:09.976
trench or an outcrop, accuracy
is on the order of 10 centimeters.

00:02:10.000 --> 00:02:13.256
The maximum range
of the scanner is 5 meters.

00:02:13.280 --> 00:02:16.240
The models are only roughly
georeferenced using the GPS

00:02:16.240 --> 00:02:19.520
of your device, but in the
3D Scanner app, this function

00:02:19.520 --> 00:02:23.176
is off by default, so make
sure that you turn it on.

00:02:23.200 --> 00:02:27.600
The total file size of a project is
less than 1 gigabyte, but the shareable

00:02:27.600 --> 00:02:31.576
textured model outputs are
generally less than 100 megabytes.

00:02:31.600 --> 00:02:36.000
A single scan is usually limited to an
area of several tens of square meters,

00:02:36.000 --> 00:02:40.616
but multiple scans can be combined
using software like CloudCompare.

00:02:40.640 --> 00:02:45.440
The sensor performs badly with
reflective surfaces, in dense vegetation,

00:02:45.440 --> 00:02:50.856
or if the complexity of the object
creates a lot of laser shadows.

00:02:50.880 --> 00:02:56.000
Comparing SfM to Lidar, SfM is
clearly the winner in image resolution,

00:02:56.000 --> 00:03:00.240
sharpness, and clarity, but for many
uses, the iDar scan quality will be

00:03:00.240 --> 00:03:05.280
good enough. It also suffers
from the same issues as SfM,

00:03:05.280 --> 00:03:08.640
where the time of day, shadows,
and lighting have a big impact

00:03:08.640 --> 00:03:12.321
on the quality of
your trench model.

00:03:13.600 --> 00:03:17.360
My own experience learning how to
use this tool has been a series of trials

00:03:17.360 --> 00:03:20.080
and errors, so I’m going to
show you several different examples

00:03:20.080 --> 00:03:23.656
of applications before getting
to our 3D trenching work.

00:03:23.680 --> 00:03:26.960
The first example is from our field
work in Kyrgyzstan this summer.

00:03:26.960 --> 00:03:30.480
Here in the Kazarman Basin,
a left-lateral strike-slip fault forms

00:03:30.480 --> 00:03:34.456
a prominent surface expression
across the landscape.

00:03:34.480 --> 00:03:38.960
This fault forms a series of grabens,
and in this area, we excavated two

00:03:38.960 --> 00:03:43.840
trenches across uphill-facing scarps.
The trench reveals coarse alluvial

00:03:43.840 --> 00:03:48.620
stratigraphy that was readily
imaged with the iOS scanner.

00:03:49.360 --> 00:03:53.520
The final image resolution is quite good,
and you can see the fine details of many

00:03:53.520 --> 00:03:57.760
gravels and cobbles. There are some
issues with duplicated cobbles,

00:03:57.760 --> 00:04:01.280
one of them circled here in red,
but the most noticeable issue

00:04:01.280 --> 00:04:04.480
is the string jumping around.
As the string is hanging in front of the

00:04:04.480 --> 00:04:08.536
trench wall, the scanner has difficulty
placing it in the correct location.

00:04:08.560 --> 00:04:12.560
This is also a frequent problem with
SfM, and I don’t recommend hanging

00:04:12.560 --> 00:04:18.221
a full string grid prior to photographing
with either of these methods.

00:04:19.280 --> 00:04:22.560
I can make measurements and
interpretations of the resulting

00:04:22.560 --> 00:04:26.640
3D trench, and here I used
Blender to make a 3D trench log.

00:04:26.640 --> 00:04:30.480
The cobbly fan layer is dropped
down by the fault, forming a graben,

00:04:30.480 --> 00:04:33.843
and has been infilled
with finer sediments.

00:04:36.080 --> 00:04:39.040
One of the places where this tool
really shines is when you are

00:04:39.040 --> 00:04:41.920
in a situation with limited
time or equipment.

00:04:41.920 --> 00:04:46.080
During an initial project kickoff visit
to Azerbaijan this fall, we were only

00:04:46.080 --> 00:04:49.520
allotted two days of field time to
do some rapid reconnaissance.

00:04:49.520 --> 00:04:53.280
Here the Iranian Plate is crushing into
the Caucasus Mountains at a rate of

00:04:53.280 --> 00:04:56.800
8 to 12 millimeters per year,
and we think most of this compression

00:04:56.800 --> 00:05:00.880
is accommodated by this massive thrust
fault. At the site I’m going to show you,

00:05:00.880 --> 00:05:05.609
a fault scarp at the front of this folded
terrace intersects a natural river bed.

00:05:06.400 --> 00:05:10.480
Due to the time constraints, we only had
about 45 minutes to spend at this site,

00:05:10.480 --> 00:05:13.200
similar to many earthquake
response situations.

00:05:13.200 --> 00:05:15.120
When we arrived,
this crew was in the process

00:05:15.120 --> 00:05:19.360
of destroying this natural fault
exposure to build a new irrigation canal,

00:05:19.360 --> 00:05:23.336
so scanning the site quickly
was our only option.

00:05:23.360 --> 00:05:26.480
Here is a preview of the model.
This model is also available

00:05:26.480 --> 00:05:30.560
on my website. The outcrop is convex,
and due to the height, scanning was

00:05:30.560 --> 00:05:34.856
difficult. Using Blender,
I can create a 3D trench log.

00:05:34.880 --> 00:05:39.953
I can also make measurements in the
model. I can then export the model.

00:05:43.520 --> 00:05:47.040
It’s important to note the difference
in 3D viewer terminology between a

00:05:47.040 --> 00:05:50.480
prospective view, where dimensions
are scaled based on distance from the

00:05:50.480 --> 00:05:55.040
camera, and an orthographic view,
where dimensions are true to reality.

00:05:55.040 --> 00:05:58.376
To make a proper trench log,
you need an orthographic view.

00:05:58.400 --> 00:06:02.320
Unfortunately, there isn’t a built-in
way in the 3D scanner app to export

00:06:02.320 --> 00:06:07.120
the 3D models into a full-resolution
2D orthoimage for making your log,

00:06:07.120 --> 00:06:11.896
but I published a tutorial on how to
do this on my website with Blender.

00:06:11.920 --> 00:06:13.976
Here’s an
exported orthoimage.

00:06:14.000 --> 00:06:18.640
On the right, sub-horizontal cobble
and silt layers rest on top of the dipping,

00:06:18.640 --> 00:06:21.360
and apparently folded,
cobble layers to the left.

00:06:21.360 --> 00:06:22.960
At the lower right
of the outcrop,

00:06:22.960 --> 00:06:26.296
the cobble layers appear to
steeply work down.

00:06:26.320 --> 00:06:30.400
I interpret this folding to be a result of
coseismic deformation but that the

00:06:30.400 --> 00:06:34.880
main fault is still buried. The horizontal
layers post-date an earthquake.

00:06:34.880 --> 00:06:38.080
The radiocarbon age of charcoals
sampled from the folded layers

00:06:38.080 --> 00:06:41.976
demonstrate that this earthquake
happened within the last 2,000 years.

00:06:42.000 --> 00:06:46.683
This is a completely usable quality
image for a publication figure.

00:06:47.440 --> 00:06:50.536
Now we’ll get to the proper
3D trenching that I promised.

00:06:50.560 --> 00:06:54.480
North of Lake Tahoe in Truckee,
California, in the northern Walker Lane

00:06:54.480 --> 00:06:58.216
is the northeast-striking
left-lateral Dog Valley Fault.

00:06:58.240 --> 00:07:01.440
This fault runs directly through
the Stampede Reservoir Dam.

00:07:01.440 --> 00:07:04.720
It produced a magnitude 6
earthquake in 1966,

00:07:04.720 --> 00:07:08.386
but no surface rupture
was observed along the fault.

00:07:09.200 --> 00:07:12.320
We excavated a trench across
this 2-meter-high fault scarp

00:07:12.320 --> 00:07:15.600
and then expanded that trench
one slice at a time to cover

00:07:15.600 --> 00:07:18.216
approximately 4 meters
along the fault.

00:07:18.240 --> 00:07:21.760
In total, we recorded
14 across-fault exposures

00:07:21.760 --> 00:07:24.696
and two fault-parallel
exposures.

00:07:24.720 --> 00:07:30.914
The across-fault exposures are
separated by about 20 to 80 centimeters.

00:07:32.800 --> 00:07:36.320
Now I will give you a tour of the
3D trench using the software program

00:07:36.320 --> 00:07:40.000
Georeka. First is the initial
fault-perpendicular cut across

00:07:40.000 --> 00:07:42.696
the scarp showing the
initial two trench walls.

00:07:42.720 --> 00:07:46.400
Then I added two fault-parallel
cuts on either side of the fault zone

00:07:46.400 --> 00:07:49.482
and two more
across-fault cuts.

00:07:50.000 --> 00:07:54.856
Next, I zoom in, and we’ll fly through
the fault zone one cut at a time.

00:07:54.880 --> 00:07:59.520
There is variability in the expression of
the fault in stratigraphy exposed in each

00:07:59.520 --> 00:08:04.961
cut, with sometimes multiple fault
strands and sometimes a single strand.

00:08:04.961 --> 00:08:08.376
I interpret that this is the
result of echelon shearing.

00:08:08.400 --> 00:08:11.520
This variability in exposure
really highlights the utility

00:08:11.520 --> 00:08:13.496
of making
multiple cuts.

00:08:13.520 --> 00:08:18.371
Alone, some of these slices would
not give us much useful information.

00:08:20.260 --> 00:08:31.840
[silence]

00:08:31.840 --> 00:08:35.656
In each cut, I can
draw the fault as a line.

00:08:35.680 --> 00:08:38.880
I can then interpolate these
fault lines into fault surfaces

00:08:38.880 --> 00:08:43.174
through multiple cuts,
here shown in translucent red.

00:08:46.880 --> 00:08:49.920
I can do the same with the upper
and lower contacts of this

00:08:49.920 --> 00:08:55.172
15-centimeter-thick channel margin
to build a 3D model of this bed.

00:08:56.160 --> 00:08:59.440
The margin of this channel sequence
makes a piercing line that I will trace

00:08:59.440 --> 00:09:02.000
to the fault zone and
use to measure the offset.

00:09:02.000 --> 00:09:06.000
For the non-paleoseismologists here,
it is often quite difficult to get offset

00:09:06.000 --> 00:09:09.828
measurements of a
strike-slip fault from trenches.

00:09:10.800 --> 00:09:14.640
Finally, I’m able to measure
115-centimeter lateral offset

00:09:14.640 --> 00:09:18.829
of the margin of the channel sequence
where it intersects the fault.

00:09:20.710 --> 00:09:30.880
[silence]

00:09:30.880 --> 00:09:35.200
I can check whether this offset is
reasonable by trying to match offset

00:09:35.200 --> 00:09:39.816
stratigraphy across sections, effectively
back-slipping the trench walls.

00:09:39.840 --> 00:09:42.880
On the left, I’ve matched two sides
of the fault that are offset about

00:09:42.880 --> 00:09:47.096
90 centimeters. The layers match
nearly perfectly across the fault.

00:09:47.120 --> 00:09:49.200
On the right I’m showing
an oblique view with

00:09:49.200 --> 00:09:53.576
the 90-centimeter measurement
between the matching layers.

00:09:53.600 --> 00:09:59.016
Again, doing the same, but this
time the fit is at 140 centimeters.

00:09:59.040 --> 00:10:03.416
Again, but this time
125 centimeters matches.

00:10:03.440 --> 00:10:06.080
Finally, another match
at 140 centimeters.

00:10:06.080 --> 00:10:10.400
So all together, there are offsets
ranging from 90 to 140 centimeters,

00:10:10.400 --> 00:10:16.080
or 115 plus or minus 25 centimeters.
This uncertainty is reasonable as it is

00:10:16.080 --> 00:10:19.440
about the same as the thickness of
most of these slices, and this result

00:10:19.440 --> 00:10:23.950
is the same as what I reconstructed
using the other 3D method.

00:10:24.640 --> 00:10:27.440
The rupture extends nearly to
the surface, so we don’t have

00:10:27.440 --> 00:10:30.480
any radiocarbon ages that
post-date the earthquake.

00:10:30.480 --> 00:10:36.136
Our age data and the clearly offset
black peat layers all fall between

00:10:36.160 --> 00:10:40.880
8,100 and 8,600 years before present.
So ultimately, we have one event

00:10:40.880 --> 00:10:45.829
with 115 centimeters of
displacement in the last 8,100 years.

00:10:47.440 --> 00:10:52.000
We received our iPad in the mail the day
before we started this trenching project,

00:10:52.000 --> 00:10:55.600
and this was before the 3D Scanner app
was released, so unfortunately,

00:10:55.600 --> 00:10:58.960
the best laser scans for this project
are these colored point clouds that

00:10:58.960 --> 00:11:02.800
we collected using the SiteScape app,
which aren’t quite as nice as

00:11:02.800 --> 00:11:05.976
the textured models
that 3D Scanner produces.

00:11:06.000 --> 00:11:09.120
So, for this project, I used
CloudCompare to combine textured

00:11:09.120 --> 00:11:15.976
model using camera-based SfM with the
spatial information from the Lidar scans.

00:11:16.000 --> 00:11:18.560
I first had to align
all of the different Lidar scans

00:11:18.560 --> 00:11:21.576
into a common
spatial reference.

00:11:21.600 --> 00:11:25.920
Then I fit each textured SfM
model to the respective scans.

00:11:25.920 --> 00:11:28.880
I used the trench flags as my
reference points when aligning

00:11:28.880 --> 00:11:32.160
the different models.
In this video, I’m showing how that

00:11:32.160 --> 00:11:36.296
fitting works, and while there is some
error, the end result is quite good.

00:11:36.320 --> 00:11:40.800
This was an incredibly time-consuming
process, and it’s difficult to convey

00:11:40.800 --> 00:11:44.696
how much information is really
contained in this 3D trench.

00:11:44.720 --> 00:11:49.336
This is a really powerful tool for
accurately scaling your SfM model.

00:11:49.360 --> 00:11:52.800
You can also directly import
your 3D texture into Agisoft

00:11:52.800 --> 00:11:55.336
and use it to
provide control points.

00:11:55.360 --> 00:11:59.040
I recommend placing high-contrast
colored flags throughout your model

00:11:59.040 --> 00:12:01.280
to use as reference points
when you’re processing

00:12:01.280 --> 00:12:04.136
the SfM and Lidar
models together.

00:12:04.160 --> 00:12:07.760
Flags that are in stable areas
also serve to tie models together

00:12:07.760 --> 00:12:11.869
when you’re progressively
excavating a 3D trench.

00:12:13.200 --> 00:12:15.520
Before I finish today,
I want to give an overview

00:12:15.520 --> 00:12:18.536
of some of the different
software packages I’ve tried.

00:12:18.560 --> 00:12:22.800
How many of you have worked with an
OBJ or other 3D textured file before?

00:12:22.800 --> 00:12:26.696
It requires a different set of tools
than you’re probably used to using.

00:12:26.720 --> 00:12:30.320
I’ve shown video so far of the
3D scanner app Georeka

00:12:30.320 --> 00:12:34.160
and CloudCompare.
Blender is a free 3D model software

00:12:34.160 --> 00:12:38.696
and quite powerful, though
the learning curve is steep.

00:12:38.720 --> 00:12:42.000
LIME is much more user-friendly
than Blender and has functions

00:12:42.000 --> 00:12:45.280
for drawing points, lines, and planes,
as well as making measurements,

00:12:45.280 --> 00:12:48.936
but it requires an
annual subscription.

00:12:48.960 --> 00:12:52.400
Sketchfab is a browser-based
viewer that is very easy to use,

00:12:52.400 --> 00:12:56.000
and the free version is useful.
You can upload directly to

00:12:56.000 --> 00:12:58.320
Sketchfab from within
the 3D Scanner app

00:12:58.320 --> 00:13:00.880
and add a few interpretive
points without paying.

00:13:00.880 --> 00:13:02.560
I uploaded this model
from the field

00:13:02.560 --> 00:13:06.150
less than 10 minutes after
starting the scanning.

00:13:07.040 --> 00:13:11.440
V3Geo is another browser-based
model viewer, similar to Sketchfab,

00:13:11.440 --> 00:13:14.720
but with a geologic focus.
It has more advanced measurement tools

00:13:14.720 --> 00:13:20.160
than Sketchfab, and they plan to add
interpretation functions in the future.

00:13:20.160 --> 00:13:24.080
In sum, iDar is a powerful and
accessible field tool that allows for

00:13:24.080 --> 00:13:26.240
collecting accurate
spatial data.

00:13:26.240 --> 00:13:32.856
For most results, it should still be
combined with GPS and SfM surveys.

00:13:32.880 --> 00:13:36.720
Aside from laser scanning and mapping
tools, you can also draw directly on

00:13:36.720 --> 00:13:41.120
your iPad using a pen in a variety
of different 2D drawing softwares.

00:13:41.120 --> 00:13:45.336
This is really useful
creating a trench log in the field.

00:13:45.360 --> 00:13:48.080
I recommend using the
3D Scanner app, Sketchfab,

00:13:48.080 --> 00:13:51.016
CloudCompare,
and LIME or Blender.

00:13:51.040 --> 00:13:54.080
MATLAB is also a useful
free software but redundant

00:13:54.080 --> 00:13:56.936
functionality
with CloudCompare.

00:13:56.960 --> 00:14:00.800
I’ve also tried Autodesk and
Adobe’s 3D modeling softwares and

00:14:00.800 --> 00:14:05.200
cannot recommend them for this use.
I didn’t find Golden Software’s Surfer

00:14:05.200 --> 00:14:08.240
or Voxler modeling
programs useful for this, either.

00:14:08.240 --> 00:14:11.600
Unfortunately, standard seismic
interpretation softwares

00:14:11.600 --> 00:14:14.856
aren’t able to use these
types of 3D models.

00:14:14.880 --> 00:14:17.840
I’m still waiting to try Moog,
but it is very expensive,

00:14:17.840 --> 00:14:21.176
so cheaper or free
options are preferable.

00:14:21.200 --> 00:14:25.576
Learning to work with 3D data is an
important new skill for geoscientists.

00:14:25.600 --> 00:14:28.960
We’re truly in the
Wild West of the 3D data era.

00:14:28.960 --> 00:14:32.160
With these – with these tools,
we’ve effectively able to digitally

00:14:32.160 --> 00:14:37.120
recreate the Earth at 1-to-1 scale.
There are many emerging technologies

00:14:37.120 --> 00:14:41.200
and applications right now, from
scanning and imaging to 3D printing,

00:14:41.200 --> 00:14:45.120
virtual reality, and augmented reality,
and it’s really exciting to see

00:14:45.120 --> 00:14:46.880
what the future
might bring.