Seeing the pattern for the pixels: Austin Troy at TEDxUVM
-
0:15 - 0:17All right, thanks Chris
and thanks for having me here. -
0:17 - 0:19Can everybody hear me OK?
-
0:22 - 0:25So, today I'm going to talk
-
0:25 - 0:27about a type of data
we're all very familiar with -
0:27 - 0:31and I think most of us like, intrinsically.
-
0:31 - 0:35And that is geographic data
and particularly imagery of the Earth. -
0:35 - 0:38And we've already seen
some examples of that today. -
0:38 - 0:42I'm going to start
with a little show and tell here. -
0:42 - 0:44I had to bring a prop, I couldn't resist.
-
0:44 - 0:51My old Macintosh Power Book 145 from 1992.
-
0:51 - 0:56This is the first computer
I had with a hard drive. -
0:56 - 0:59It came with 4, was it 4?
-
0:59 - 1:04Actually with about 6 megabytes of memory.
-
1:04 - 1:07That was a big deal
and I was just blown away. -
1:07 - 1:09I couldn't believe I had that much memory.
-
1:09 - 1:11I'm just going to put this back here.
-
1:11 - 1:15I couldn't believe
I had that much memory at my fingertips. -
1:15 - 1:17Today, we now all have computers.
-
1:17 - 1:19I can go down to Staples
and buy a computer -
1:19 - 1:22that has a quarter million times
more memory than that -
1:22 - 1:25for about $400 or $600
or something like that. -
1:25 - 1:28Times have changed
and that's 20 years ago. -
1:28 - 1:32As a result,
with all this increased computing power, -
1:32 - 1:34we are drowning in data.
-
1:34 - 1:36We're just absolutely drowning in data.
-
1:36 - 1:39And one of the types of data
we are most drowning in -
1:39 - 1:42is remote sensing
or imagery data, satellite data, -
1:42 - 1:44aerial data, things like that.
-
1:44 - 1:46And we've all played around
with this, I'm sure. -
1:46 - 1:49We all love Google Earth,
it's free and it's fun. -
1:49 - 1:52And it's just teaming with imagery.
-
1:52 - 1:54So, what do we do with all this stuff?
-
1:54 - 1:56How do we make use of this?
-
1:56 - 1:59Here's an image of Baltimore.
This is urban Baltimore. -
1:59 - 2:01It's got all these great objects.
-
2:01 - 2:03I can look in there and see,
-
2:03 - 2:04it's hard with this projector,
-
2:04 - 2:07but I can see trees and buildings,
things like that. -
2:07 - 2:08And let's just say
-
2:08 - 2:12I wanted to actually do some kind
of a quantitative study with that. -
2:12 - 2:14Say I had to do something
that required knowing -
2:14 - 2:16where the trees really were.
-
2:16 - 2:17I can see where the trees are,
-
2:17 - 2:20but the computer doesn't know,
it has no clue what a tree is. -
2:20 - 2:23Let's just say I wanted to do
something like, -
2:23 - 2:25these are actually
the locations of crimes, -
2:25 - 2:29say I wanted to know
if the density of trees affects crime. -
2:29 - 2:32There's no way I can do that
with imagery the way we have it now, -
2:32 - 2:36in a computing environment.
-
2:36 - 2:39Part of the reason for this
-
2:39 - 2:41is that computers
aren't really good at recognizing things -
2:41 - 2:43the way that we can recognise things.
-
2:43 - 2:48We are excellent are recognizing
things with very slight differences. -
2:48 - 2:51I can tell you within two seconds
-
2:51 - 2:53that that's George Carlin
and that's Sigmund Freud. -
2:53 - 2:56That that is the Big Lebowski
and that is Eddie Vedder. -
2:56 - 3:00For me to train a computer
to recognise the difference -
3:00 - 3:04between the Big Lebowski, a.k.a. the Dude
and Eddie Vedder, -
3:04 - 3:07would take me unbelievable amounts
of time to do. -
3:07 - 3:12Yet I can do that instantly,
so that's an issue right here. -
3:12 - 3:15So, let's cut to the chase here.
-
3:15 - 3:18On the left I have raw data,
-
3:18 - 3:21color infrared remote sense imagery.
-
3:21 - 3:27On the right I have
a classified GIS layer. -
3:27 - 3:30That is usable information.
-
3:30 - 3:34The computer knows what's grass,
what's buildings and knows what's trees. -
3:34 - 3:37How do I get from one to the other?
-
3:37 - 3:41This is a major major conundrum
in today's world of high resolution data. -
3:41 - 3:44Here's an image
of just a typical suburban area. -
3:44 - 3:46I look at it and I see
all sort of features -
3:46 - 3:48and I see that it is at a
very fine resolution. -
3:48 - 3:52If I were to be working
with remote sensing data 15 years ago, -
3:52 - 3:55I'd have coarse resolution imagery.
-
3:55 - 3:57This is the same exact location
using 30 meter pixels. -
3:57 - 4:02Back then, classifying this stuff
was a qualitatively different thing, -
4:02 - 4:06because all I really needed to do
is get in the general ball park. -
4:06 - 4:08These pixels here
are sort of generally urbanized, -
4:08 - 4:11these pixels here are generally forested.
-
4:11 - 4:15I didn't really have to know
about the specific identity of objects. -
4:15 - 4:19Now fast forward to today
and I've got imagery which I can zoom in -
4:19 - 4:23and I can see
a million different types of objects. -
4:23 - 4:30From cars in a parking lot
to shipping containers, -
4:30 - 4:34to the cranes
at the shipping containers facility -
4:34 - 4:38to the mechanicals on top of a building
-
4:38 - 4:39to, I love this one,
-
4:39 - 4:43this is the Sphinx
at the Luxor Hotel in Las Vegas. -
4:43 - 4:45Try telling a computer what that is.
-
4:45 - 4:49Here's another Vegas one,
I love Google Earth in Vegas, -
4:49 - 4:50it's the best, it's so much fun.
-
4:50 - 4:53This is a tropical fish-shaped pool.
-
4:53 - 4:56Again, not so easy to tell a computer
what that is. -
4:56 - 4:58And this is the best of all.
-
4:58 - 5:03I believe this is high-resolution
satellite imagery -
5:03 - 5:07of camels in the middle of Africa
and it's part of this Google, -
5:07 - 5:12National Geographic,
Africa mega fly-over project. -
5:12 - 5:14The number of possible things
-
5:14 - 5:16I have to prepare computers
to be ready for, -
5:16 - 5:19the types of objects
out on the surface of the Earth -
5:19 - 5:20surface of the Earth,
is staggering. -
5:21 - 5:25And it's a project
that we'll never see finished, -
5:25 - 5:29that project of teaching computers
the artificial intelligence, -
5:29 - 5:32giving them the artificial intelligence
they need -
5:32 - 5:34to recognize all of this variation
on the Earth. -
5:34 - 5:35Here's another even better one.
-
5:35 - 5:38This is actually a real thing,
this is the Colonel. -
5:38 - 5:43Someone actually did mega art
of the Colonel in the middle of Nevada. -
5:43 - 5:46It's interesting
how many of these come from Nevada. -
5:46 - 5:49(Laughter)
-
5:49 - 5:52So, let's explain why it's difficult
-
5:52 - 5:55to use the methods
we've always used in the past -
5:55 - 5:58for this generation
of high resolution imagery. -
5:58 - 6:01Here's a high resolution image
of Burlington. -
6:01 - 6:06If I try to classify each pixel,
pixel by pixel, -
6:06 - 6:11I get this awful pixelated
meaningless gobbledygook. -
6:11 - 6:14If I look at a particular object
like a house, that house is made up, -
6:14 - 6:17it's hard to see, but it's just dozens
of different pixel values -
6:17 - 6:19that don't really mean anything.
-
6:19 - 6:24If I take a single tree, again,
is made up of a gobbledygook of pixels. -
6:24 - 6:27Now, if I zoom in on that tree,
for instance, -
6:27 - 6:33I will see that it's made up of pixels,
lots of different tones, different colors, -
6:33 - 6:38and if this is the direct representation
in classified pixels, -
6:38 - 6:39it's meaningless, right?
-
6:39 - 6:41This is not finding objects.
-
6:41 - 6:46So, I need to teach a computer
to see objects and to think like me. -
6:46 - 6:49So this means teaching a computer
to think like a human, -
6:49 - 6:52which means working
based on shape size, tone, pattern, -
6:52 - 6:55texture, site and association,
a lot of this is all spatial. -
6:55 - 7:00We have to stop thinking pixel by pixel
and start thinking of things spatially. -
7:00 - 7:04That means taking an image
and, what's called segmenting it, -
7:04 - 7:06turning it into objects.
-
7:06 - 7:09And the process of segmenting it
is very difficult. -
7:09 - 7:13You have to train a computer
to segment imagery correctly. -
7:13 - 7:15And if I look, here is a house,
there's one side of the roof -
7:15 - 7:18and another side of the roof,
there is the driveway, -
7:18 - 7:20they're segmented as different objects
-
7:20 - 7:22and I can then re-aggregate those objects
-
7:22 - 7:25into something that is just a house
and another one that is just a driveway. -
7:25 - 7:28And at the end of the day,
what I'm going to end up with -
7:28 - 7:29is something like this.
-
7:29 - 7:32I will be able to tell you the difference.
-
7:32 - 7:38Even though the spectral signature
is the same of this roof and this road, -
7:38 - 7:41I know that their compactness
factor is different, -
7:41 - 7:44and because of that,
because of the shape metrics of them, -
7:44 - 7:46I can tell you which one's a roof
and which one's a road, -
7:46 - 7:49and I can start
classifying things in that way. -
7:53 - 7:56To do this requires huge rule sets.
-
7:56 - 8:01The rule sets could be dozens and dozens
to over a hundred pages long -
8:01 - 8:06of all these classification rules
and I won't bore you with the details. -
8:06 - 8:10I also will make use of
a lot of ancillary data. -
8:10 - 8:13There's all sort of great GIS data
that helps me classify things now. -
8:13 - 8:16Most cities are collecting things
about building footprints, -
8:16 - 8:19we know where parcels are,
we know where sewer lines are -
8:19 - 8:20and roads are and things like that.
-
8:20 - 8:22We can use this to help us,
-
8:22 - 8:24but the most important
form of ancillary data -
8:24 - 8:28that's out there today is called LIDAR:
Light Detection and Ranging. -
8:28 - 8:30And LIDAR has been used
in engineering for a while -
8:30 - 8:33and it allows us to essentially create
models of the surface. -
8:33 - 8:36This is Columbus Circle
in Central Park in New York City -
8:36 - 8:39and this is a surface elevation
of the trees. -
8:39 - 8:42The LIDAR tells me
where the canopy of the trees is, -
8:42 - 8:48where the tops of the buildings are,
where the ground surface is too. -
8:48 - 8:51And I can create these incredibly detailed
models of the world -
8:51 - 8:52so now I'm not just working
-
8:52 - 8:56with spatial spectral information,
reflectance information, -
8:56 - 8:59I'm also working with height information,
I know the heights of things -
8:59 - 9:02so I can see two objects
that are green and woody, -
9:02 - 9:06but I can tell that one of them
is a shrub and one of them is a tree. -
9:06 - 9:09And this is just zooming
into that stuff there. -
9:09 - 9:12Now the problem is,
this is incredibly data-intensive -
9:12 - 9:18and nobody's figured out until recently,
and I mean like maybe two years ago, -
9:18 - 9:21people were doing this
on a tile by tile basis -
9:21 - 9:23to work on one little tile of data
at a time -
9:23 - 9:25that might be, you know,
-
9:25 - 9:29one, that just that red outline
that you see right there, -
9:29 - 9:31and that might be half a gigabyte
or something like that. -
9:31 - 9:35So we've worked on turning this
into an enterprise environment, -
9:35 - 9:38that's what we have to do,
make an enterprise environment out of this -
9:38 - 9:41so we can start looking
at thousands of tiles of data at a time, -
9:41 - 9:43and we've successfully done that.
-
9:44 - 9:48My lab, which is the
Spatial Analysis Lab. -
9:48 - 9:52The Spatial Analysis lab is the lab I run,
-
9:52 - 9:55and they've been doing this stuff
for a number of years, -
9:55 - 9:59and they've collected,
through 64 projects, 837 communities, -
9:59 - 10:04covering 28 million people,
almost 9000 square miles of data mapped, -
10:04 - 10:07250 billion pixels of land cover
products generated -
10:07 - 10:09and 110 terabytes of data.
-
10:09 - 10:12So this is a major undertaking
but it's only the beginning. -
10:12 - 10:16Going back to the crime data
that I was telling you about those trees, -
10:16 - 10:17so here's Baltimore again.
-
10:17 - 10:20Using this method,
we turn data into information. -
10:20 - 10:23We get trees,
we now know where trees are. -
10:23 - 10:24I overlay it with the crime.
-
10:24 - 10:28I end up with information,
I can now do a study, -
10:29 - 10:31and we just submitted this
for publication. -
10:31 - 10:34We just found out in fact
there's a strong negative correlation -
10:34 - 10:35between trees and crime
-
10:35 - 10:38even when adjust for about
fifty other things. -
10:38 - 10:43We couldn't have done that
without this sort of information. -
10:43 - 10:45So with that, I will say thanks
to the people -
10:45 - 10:47from the Spatial Analysis Lab,
-
10:47 - 10:50and particularly Jarlath O'Neil-Dunne
who helped me put this together -
10:50 - 10:53and has been doing this research
for a long time and thanks to you. -
10:53 - 10:56Thank you. (Applause)
- Title:
- Seeing the pattern for the pixels: Austin Troy at TEDxUVM
- Description:
-
Austin Troy's work focuses on land use – particularly the causes and impacts of urban/suburban development and the effectiveness of policies in mediating those impacts. He is also interested in studying the role and importance of environmental assets in urban systems and in quantifying the spatial distribution of ecosystem services.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDxTalks
- Duration:
- 10:58
Krystian Aparta edited English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Krystian Aparta approved English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Krystian Aparta commented on English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Krystian Aparta edited English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Krystian Aparta edited English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Krystian Aparta edited English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Krystian Aparta edited English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM | ||
Denise RQ accepted English subtitles for Seeing the pattern for the pixels: Austin Troy at TEDxUVM |
Krystian Aparta
Note: in the new editor, you can see the character length of each subtitle, as well as its reading speed (characters/second). For languages based on the Latin alphabet, the maximum subtitle length is 84 characters (subtitles over 42 characters need to be broken into two lines). The maximum reading speed should be less than 22 characters per second. You can access the new editor by clicking "Beta: Save and open in new editor" after opening the task in the old interface. To learn more about line length, line breaking and reading speed, watch this tutorial: http://www.youtube.com/watch?v=yvNQoD32Qqo //////////////////////////////////////////////////////////////////////////////////////
Don't capitalize every word in the talk title. The correct TEDx title format is [Talk title]: [Speaker name] at [Event name]. See http://translations.ted.org/wiki/How_to_Tackle_a_Transcript#Title_and_description_standard //////////////////////////////////////////////////////////////////////////////////////
I fixed the reading speed of the subtitles where it was over 21 characters per second. I did this by either compressing the text (see http://translations.ted.org/wiki/How_to_Compress_Subtitles) or by editing the timing of the subtitle. In some cases, I merged subtitles to create a bigger subtitle with the correct reading speed. In order to merge subtitles, copy the text of the second subtitle, delete the second subtitle, paste its text into the first subtitle and extended its time to cover the duration of the deleted subtitle. To learn more about line length, line breaking and reading speed, watch this tutorial: http://www.youtube.com/watch?v=yvNQoD32Qqo //////////////////////////////////////////////////////////////////////////////////////
Parentheses are used for sound representation (e.g. (Applause)). Please do not use parentheses for other purposes. //////////////////////////////////////////////////////////////////////////////////////
The duration of a subtitle should not be over 7 seconds. I split some subtitles whose duration extended that limit (to split a subtitle, you can shorten the duration of the current subtitle and insert another subtitle into the resulting "gap").
//////////////////////////////////////////////////////////////////////////////////////
The maximum length of a subtitle is 84. I corrected the subtitles that were over this limit. //////////////////////////////////////////////////////////////////////////////////////
Generally, please don't include slips of the tongue and mid-sentence changes that do not alter the meaning of the whole sentence. For example, if the speaker says "I can see that there are some-- I can see some buildings here," just transcribe it as "I can see some buildings here" or "I can see that there are some buildings here," depending on which version would be more suitable in terms of reading speed. An example from this transcript would be "and I can see objects that are, a million different types of objects" --> "and I can see a million different types of objects". //////////////////////////////////////////////////////////////////////////////////////
Please be careful about proper names. "and particularly to the ones who helped me put this together" should read "and particularly Jarlath O'Neil-Dunne who helped me put this together." Jarlath O'Neil-Dunne is the Director of the Spatial Analysis Lab, something that I looked up by going to their website. If you can't make out a word or phrase, please ask for help in https://www.facebook.com/groups/ITranscribeTEDxtalks/ and if you can't find any help there, you can put "[unclear]" in the subtitles where you can't make out a word, so that it's easier on the reviewer and approver.