Dan Berkenstock: The world is one big dataset. Now, how to photograph it ...

61,716 views ・ 2014-02-04

TED


Please double-click on the English subtitles below to play the video.

00:12
Five years ago, I was a Ph.D. student
0
12498
2599
00:15
living two lives.
1
15097
1644
00:16
In one, I used NASA supercomputers
2
16741
2316
00:19
to design next-generation spacecraft,
3
19057
2728
00:21
and in the other I was a data scientist
4
21785
2579
00:24
looking for potential smugglers
5
24364
2419
00:26
of sensitive nuclear technologies.
6
26783
3239
00:30
As a data scientist, I did a lot of analyses,
7
30022
2859
00:32
mostly of facilities,
8
32881
1467
00:34
industrial facilities around the world.
9
34348
2271
00:36
And I was always looking for a better canvas
10
36619
2688
00:39
to tie these all together.
11
39307
2066
00:41
And one day, I was thinking about how
12
41373
1866
00:43
all data has a location,
13
43239
2124
00:45
and I realized that the answer
14
45363
1733
00:47
had been staring me in the face.
15
47096
1888
00:48
Although I was a satellite engineer,
16
48984
2638
00:51
I hadn't thought about using satellite imagery
17
51622
3171
00:54
in my work.
18
54793
1566
00:56
Now, like most of us, I'd been online,
19
56359
1947
00:58
I'd see my house, so I thought,
20
58306
1782
01:00
I'll hop in there and I'll start looking up
21
60088
2298
01:02
some of these facilities.
22
62386
1417
01:03
And what I found really surprised me.
23
63803
1969
01:05
The pictures that I was finding
24
65772
1674
01:07
were years out of date,
25
67446
1901
01:09
and because of that,
26
69347
1068
01:10
it had relatively little relevance
27
70415
1671
01:12
to the work that I was doing today.
28
72086
2460
01:14
But I was intrigued.
29
74546
1461
01:16
I mean, satellite imagery is pretty amazing stuff.
30
76007
3236
01:19
There are millions and millions of sensors
31
79243
2303
01:21
surrounding us today,
32
81546
1444
01:22
but there's still so much we don't know on a daily basis.
33
82990
2960
01:25
How much oil is stored in all of China?
34
85950
3753
01:29
How much corn is being produced?
35
89703
2933
01:32
How many ships are in all of our world's ports?
36
92636
4071
01:36
Now, in theory, all of these questions
37
96707
2559
01:39
could be answered by imagery,
38
99266
2244
01:41
but not if it's old.
39
101510
1727
01:43
And if this data was so valuable,
40
103237
2458
01:45
then how come I couldn't get my hands
41
105695
1932
01:47
on more recent pictures?
42
107627
2550
01:50
So the story begins over 50 years ago
43
110177
3135
01:53
with the launch of the first generation
44
113312
1871
01:55
of U.S. government photo reconnaissance satellites.
45
115183
3600
01:58
And today, there's a handful
46
118783
1696
02:00
of the great, great grandchildren
47
120479
2195
02:02
of these early Cold War machines
48
122674
1665
02:04
which are now operated by private companies
49
124339
2009
02:06
and from which the vast majority of satellite imagery
50
126348
2709
02:09
that you and I see on a daily basis comes.
51
129057
2825
02:11
During this period, launching things into space,
52
131882
2914
02:14
just the rocket to get the satellite up there,
53
134796
2677
02:17
has cost hundreds of millions of dollars each,
54
137473
4601
02:22
and that's created tremendous pressure
55
142074
1752
02:23
to launch things infrequently
56
143826
2289
02:26
and to make sure that when you do,
57
146115
1344
02:27
you cram as much functionality in there as possible.
58
147459
3639
02:31
All of this has only made satellites
59
151098
1611
02:32
bigger and bigger and bigger
60
152709
2417
02:35
and more expensive,
61
155126
1598
02:36
now nearly a billion, with a b, dollars per copy.
62
156724
5037
02:41
Because they are so expensive,
63
161761
1699
02:43
there aren't very many of them.
64
163460
1467
02:44
Because there aren't very many of them,
65
164927
1345
02:46
the pictures that we see on a daily basis
66
166272
2456
02:48
tend to be old.
67
168728
1504
02:50
I think a lot of people actually understand this anecdotally,
68
170232
3282
02:53
but in order to visualize just how sparsely
69
173514
2539
02:56
our planet is collected,
70
176053
1884
02:57
some friends and I put together a dataset
71
177937
2066
03:00
of the 30 million pictures that have been gathered
72
180003
2587
03:02
by these satellites between 2000 and 2010.
73
182590
3484
03:06
As you can see in blue, huge areas of our world
74
186074
2755
03:08
are barely seen, less than once a year,
75
188829
2790
03:11
and even the areas that are seen most frequently,
76
191619
2272
03:13
those in red, are seen at best once a quarter.
77
193891
3748
03:17
Now as aerospace engineering grad students,
78
197639
2916
03:20
this chart cried out to us as a challenge.
79
200555
3428
03:23
Why do these things have to be so expensive?
80
203983
3455
03:27
Does a single satellite really have to cost
81
207438
2735
03:30
the equivalent of three 747 jumbo jets?
82
210173
4763
03:34
Wasn't there a way to build a smaller,
83
214936
2163
03:37
simpler, new satellite design that could enable
84
217099
3172
03:40
more timely imaging?
85
220271
2018
03:42
I realize that it does sound a little bit crazy
86
222289
3451
03:45
that we were going to go out and just
87
225740
1511
03:47
begin designing satellites,
88
227251
1847
03:49
but fortunately we had help.
89
229098
1963
03:51
In the late 1990s, a couple of professors
90
231061
2935
03:53
proposed a concept for radically reducing the price
91
233996
3481
03:57
of putting things in space.
92
237477
1772
03:59
This was hitchhiking small satellites
93
239249
2111
04:01
alongside much larger satellites.
94
241360
3016
04:04
This dropped the cost of putting objects up there
95
244376
2909
04:07
by over a factor of 100,
96
247285
1998
04:09
and suddenly we could afford to experiment,
97
249283
3086
04:12
to take a little bit of risk,
98
252369
1468
04:13
and to realize a lot of innovation.
99
253837
2266
04:16
And a new generation of engineers and scientists,
100
256103
3089
04:19
mostly out of universities,
101
259192
1599
04:20
began launching these very small,
102
260791
2343
04:23
breadbox-sized satellites called CubeSats.
103
263134
2591
04:25
And these were built with electronics obtained
104
265725
2403
04:28
from RadioShack instead of Lockheed Martin.
105
268128
3898
04:32
Now it was using the lessons learned from these early missions
106
272026
2840
04:34
that my friends and I began a series of sketches
107
274866
2695
04:37
of our own satellite design.
108
277561
1689
04:39
And I can't remember a specific day
109
279250
2784
04:42
where we made a conscious decision
110
282034
1860
04:43
that we were actually going to go out and build these things,
111
283894
2635
04:46
but once we got that idea in our minds
112
286529
2430
04:48
of the world as a dataset,
113
288959
2161
04:51
of being able to capture millions of data points
114
291120
2490
04:53
on a daily basis describing the global economy,
115
293610
3053
04:56
of being able to unearth billions of connections
116
296663
2520
04:59
between them that had never before been found,
117
299183
3165
05:02
it just seemed boring
118
302348
1648
05:03
to go work on anything else.
119
303996
2907
05:06
And so we moved into a cramped,
120
306903
2973
05:09
windowless office in Palo Alto,
121
309876
2979
05:12
and began working to take our design
122
312855
1967
05:14
from the drawing board into the lab.
123
314822
2938
05:17
The first major question we had to tackle
124
317760
2658
05:20
was just how big to build this thing.
125
320418
2334
05:22
In space, size drives cost,
126
322752
3183
05:25
and we had worked with these very small,
127
325935
2013
05:27
breadbox-sized satellites in school,
128
327948
2094
05:30
but as we began to better understand the laws of physics,
129
330042
2551
05:32
we found that the quality of pictures
130
332593
2122
05:34
those satellites could take was very limited,
131
334715
3084
05:37
because the laws of physics dictate
132
337799
1704
05:39
that the best picture you can take through a telescope
133
339503
3106
05:42
is a function of the diameter of that telescope,
134
342609
2239
05:44
and these satellites had a very small,
135
344848
1662
05:46
very constrained volume.
136
346510
1706
05:48
And we found that the best picture we would
137
348216
1795
05:50
have been able to get looked something like this.
138
350011
2485
05:52
Although this was the low-cost option,
139
352496
2245
05:54
quite frankly it was just too blurry
140
354741
1673
05:56
to see the things that make satellite imagery valuable.
141
356414
3569
05:59
So about three or four weeks later,
142
359983
2146
06:02
we met a group of engineers randomly
143
362129
2817
06:04
who had worked on the first
144
364946
1641
06:06
private imaging satellite ever developed,
145
366587
2441
06:09
and they told us that back in the 1970s,
146
369028
2118
06:11
the U.S. government had found a powerful
147
371146
2182
06:13
optimal tradeoff --
148
373328
1643
06:14
that in taking pictures at right about one meter resolution,
149
374971
3088
06:18
being able to see objects one meter in size,
150
378059
2884
06:20
they had found that they could not just get very high-quality images,
151
380943
2691
06:23
but get a lot of them at an affordable price.
152
383634
2893
06:26
From our own computer simulations,
153
386527
1976
06:28
we quickly found that one meter really was
154
388503
2005
06:30
the minimum viable product
155
390508
1764
06:32
to be able to see the drivers of our global economy,
156
392272
3079
06:35
for the first time, being able to count
157
395351
1561
06:36
the ships and cars and shipping containers and trucks
158
396912
2721
06:39
that move around our world on a daily basis,
159
399633
2824
06:42
while conveniently still not being able to see individuals.
160
402457
3804
06:46
We had found our compromise.
161
406261
1621
06:47
We would have to build something larger
162
407882
1604
06:49
than the original breadbox,
163
409486
1563
06:51
now more like a mini-fridge,
164
411049
1651
06:52
but we still wouldn't have to build a pickup truck.
165
412700
2761
06:55
So now we had our constraint.
166
415461
2624
06:58
The laws of physics dictated
167
418085
1757
06:59
the absolute minimum-sized telescope that we could build.
168
419842
3324
07:03
What came next was making the rest of the satellite
169
423166
3054
07:06
as small and as simple as possible,
170
426220
1741
07:07
basically a flying telescope with four walls
171
427961
2652
07:10
and a set of electronics smaller than a phone book
172
430613
2941
07:13
that used less power than a 100 watt lightbulb.
173
433554
3438
07:16
The big challenge became actually taking
174
436992
1900
07:18
the pictures through that telescope.
175
438892
2198
07:21
Traditional imaging satellites use a line scanner,
176
441090
3044
07:24
similar to a Xerox machine,
177
444134
1761
07:25
and as they traverse the Earth, they take pictures,
178
445895
2400
07:28
scanning row by row by row
179
448295
2196
07:30
to build the complete image.
180
450491
1675
07:32
Now people use these because they get a lot of light,
181
452166
3004
07:35
which means less of the noise you see
182
455170
1831
07:37
in a low-cost cell phone image.
183
457016
2736
07:39
The problem with them is they require
184
459752
2361
07:42
very sophisticated pointing.
185
462113
2240
07:44
You have to stay focused on a 50-centimeter target
186
464353
2546
07:46
from over 600 miles away
187
466899
1752
07:48
while moving at more than seven kilometers a second,
188
468651
2232
07:50
which requires an awesome degree of complexity.
189
470883
2749
07:53
So instead, we turned to a new generation of video sensors,
190
473632
3437
07:57
originally created for use in night vision goggles.
191
477069
3091
08:00
Instead of taking a single, high quality image,
192
480160
2899
08:03
we could take a videostream
193
483059
1387
08:04
of individually noisier frames,
194
484446
2923
08:07
but then we could recombine
195
487369
1744
08:09
all of those frames together
196
489113
1466
08:10
into very high-quality images
197
490579
1996
08:12
using sophisticated pixel processing techniques
198
492575
2593
08:15
here on the ground,
199
495168
1521
08:16
at a cost of one one hundredth a traditional system.
200
496689
3137
08:19
And we applied this technique
201
499826
1153
08:20
to many of the other systems on the satellite as well,
202
500979
2974
08:23
and day by day, our design evolved
203
503953
2840
08:26
from CAD to prototypes
204
506793
3489
08:30
to production units.
205
510282
2888
08:33
A few short weeks ago,
206
513170
1657
08:34
we packed up SkySat 1,
207
514827
1851
08:36
put our signatures on it,
208
516678
1773
08:38
and waved goodbye for the last time on Earth.
209
518451
2223
08:40
Today, it's sitting in its final launch configuration
210
520674
3481
08:44
ready to blast off in a few short weeks.
211
524155
3061
08:47
And soon, we'll turn our attention to launching
212
527216
2342
08:49
a constellation of 24 or more of these satellites
213
529558
3031
08:52
and beginning to build the scalable analytics
214
532589
2550
08:55
that will allow us to unearth the insights
215
535139
2339
08:57
in the petabytes of data we will collect.
216
537478
3189
09:00
So why do all of this? Why build these satellites?
217
540667
3973
09:04
Well, it turns out imaging satellites
218
544640
2611
09:07
have a unique ability to provide global transparency,
219
547251
3710
09:10
and providing that transparency on a timely basis
220
550961
2626
09:13
is simply an idea whose time has come.
221
553587
3411
09:16
We see ourselves as pioneers of a new frontier,
222
556998
3523
09:20
and beyond economic data,
223
560521
1764
09:22
unlocking the human story, moment by moment.
224
562285
3601
09:25
For a data scientist
225
565886
1563
09:27
that just happened to go to space camp as a kid,
226
567449
2750
09:30
it just doesn't get much better than that.
227
570199
2715
09:32
Thank you.
228
572914
2215
09:35
(Applause)
229
575129
4134
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7