1
00:00:00,630 --> 00:00:05,100
- We are happy to introduce Andrew Logan

2
00:00:05,100 --> 00:00:10,100
and his DEF CON 30 and his parents,

3
00:00:10,500 --> 00:00:12,270
he wanted me to mention his parents,

4
00:00:12,270 --> 00:00:14,160
we're happy to have him here.

5
00:00:14,160 --> 00:00:16,650
He will be talking
about ghost helicopters.

6
00:00:16,650 --> 00:00:19,962
So please give a big round
of applause for Andrew Logan.

7
00:00:19,962 --> 00:00:21,402
(audience applauding)

8
00:00:21,402 --> 00:00:22,302
The stage is yours.

9
00:00:22,302 --> 00:00:25,350
(audience applauding)

10
00:00:25,350 --> 00:00:28,590
- Thank you all so much, it is an honor,

11
00:00:28,590 --> 00:00:32,433
and I'm so thrilled to be here
at DEF CON, my first DEF CON.

12
00:00:33,540 --> 00:00:34,530
So I wanna talk to you about

13
00:00:34,530 --> 00:00:38,340
tracking the military ghost
helicopters of Washington, D.C.

14
00:00:38,340 --> 00:00:41,910
And I'll start by taking you
on a tour of Washington, D.C

15
00:00:41,910 --> 00:00:44,220
where you can see the Army National Guard

16
00:00:44,220 --> 00:00:46,560
practicing their rotor wash techniques

17
00:00:46,560 --> 00:00:48,450
over peaceful protestors

18
00:00:48,450 --> 00:00:52,110
or the Park Police practicing
the same techniques

19
00:00:52,110 --> 00:00:55,440
on ice skaters on the National Mall.

20
00:00:55,440 --> 00:00:57,480
Maybe you wanna spend a quiet day inside.

21
00:00:57,480 --> 00:00:59,700
Well, if it's around inauguration,

22
00:00:59,700 --> 00:01:02,220
you can enjoy the Department of Energy

23
00:01:02,220 --> 00:01:04,260
sweeping for background radiation

24
00:01:04,260 --> 00:01:09,260
at 150 feet above your dwelling,
back and forth, all day.

25
00:01:09,900 --> 00:01:11,520
And if that wasn't enough,

26
00:01:11,520 --> 00:01:14,490
we also have a top secret Army mission

27
00:01:14,490 --> 00:01:17,580
and the Night Vision and
Electronic Sensors Directorate

28
00:01:17,580 --> 00:01:19,680
is based in nearby Fort Belvoir,

29
00:01:19,680 --> 00:01:21,933
so they stay pretty busy even at night.

30
00:01:22,800 --> 00:01:25,410
These are just a few of
the helicopter operations

31
00:01:25,410 --> 00:01:29,880
over the District, but so
began my helicopter obsession,

32
00:01:29,880 --> 00:01:33,780
kept up late at night, finding
little historical information

33
00:01:33,780 --> 00:01:37,800
the following day but finding
a lot of Twitter users

34
00:01:37,800 --> 00:01:41,310
with the same questions who
could not identify a helicopter

35
00:01:41,310 --> 00:01:42,213
if they saw one.

36
00:01:43,680 --> 00:01:46,440
When I looked at libraries
that would help me work

37
00:01:46,440 --> 00:01:48,630
with aircraft data and Twitter,

38
00:01:48,630 --> 00:01:50,340
Python was an obvious choice.

39
00:01:50,340 --> 00:01:53,100
There are many flavors
of Python, as you know

40
00:01:53,100 --> 00:01:58,100
but as a novice coder, I
decided to go with Bad Python.

41
00:01:58,350 --> 00:02:02,074
Some of the tenants of Bad
Python use an outdated version.

42
00:02:02,074 --> 00:02:03,120
(audience laughing)

43
00:02:03,120 --> 00:02:04,605
God, I shouldn't tell you guys this.

44
00:02:04,605 --> 00:02:06,600
(Andrew laughing)

45
00:02:06,600 --> 00:02:09,033
Command line editor is always good.

46
00:02:09,960 --> 00:02:13,080
Don't consistently name your
variables and functions.

47
00:02:13,080 --> 00:02:15,720
Leave your API keys in every file

48
00:02:15,720 --> 00:02:19,380
and an absolute minimum of white space.

49
00:02:19,380 --> 00:02:22,503
This does help when you're
debugging on your phone.

50
00:02:24,210 --> 00:02:27,540
And with that, I built a basic ADS-B bot,

51
00:02:27,540 --> 00:02:30,540
not dissimilar from the famous Elon Jet,

52
00:02:30,540 --> 00:02:34,560
originally using free
data from OpenSky API

53
00:02:34,560 --> 00:02:37,140
and mapping with Matplotlib.

54
00:02:37,140 --> 00:02:39,690
Now, my wife has a master's in geography

55
00:02:39,690 --> 00:02:41,190
and when I asked her what she thought

56
00:02:41,190 --> 00:02:44,997
of this map on the left,
she said, "Looks good."

57
00:02:46,170 --> 00:02:48,660
She's very generous person.

58
00:02:48,660 --> 00:02:53,580
So now we're using Google
Map Static API custom icons

59
00:02:53,580 --> 00:02:57,420
and we use data from ADSBExchange,
which is much better.

60
00:02:57,420 --> 00:02:59,850
I highly recommend ADSBExchange.

61
00:02:59,850 --> 00:03:01,620
Anybody in the audience knows them.

62
00:03:01,620 --> 00:03:03,990
They don't filter military

63
00:03:03,990 --> 00:03:07,740
and they don't filter
private operators for profit.

64
00:03:07,740 --> 00:03:11,370
And a little background on
ADS-B, if you don't know,

65
00:03:11,370 --> 00:03:14,130
it's a transponder protocol
that has been required

66
00:03:14,130 --> 00:03:18,360
in many of the busier air
spaces since early 2020,

67
00:03:18,360 --> 00:03:20,640
and it's part of FAA NextGen plan

68
00:03:20,640 --> 00:03:24,000
to pack more flights into smaller spaces.

69
00:03:24,000 --> 00:03:28,470
It stands for Automatic
Dependent Surveillance Broadcast.

70
00:03:28,470 --> 00:03:30,270
Automatic in that it does not require

71
00:03:30,270 --> 00:03:34,320
the pilot's intervention,
dependent on a number of sensors,

72
00:03:34,320 --> 00:03:38,790
surveillance in that it reports
those sensors via broadcast

73
00:03:38,790 --> 00:03:43,740
on 978 or 1090 megahertz,
unencrypted for reception

74
00:03:43,740 --> 00:03:45,750
by ground stations like us.

75
00:03:45,750 --> 00:03:48,540
Well, it's actually intended
for reception by the tower

76
00:03:48,540 --> 00:03:52,110
and rebroadcast to other
planes, but we can receive it.

77
00:03:52,110 --> 00:03:55,830
So at the bottom, we have
a chart on the evolution

78
00:03:55,830 --> 00:03:58,470
of transponder codes
and their capabilities

79
00:03:58,470 --> 00:04:00,750
as taken from this DOD

80
00:04:00,750 --> 00:04:03,510
and Government
Accountability Office report

81
00:04:03,510 --> 00:04:07,593
on the risks to military
using these transponders.

82
00:04:08,460 --> 00:04:11,640
So it's no surprise also in this report

83
00:04:11,640 --> 00:04:14,640
you'll find the Code
of Federal Regulations

84
00:04:14,640 --> 00:04:19,400
exemption from transponder
use for agencies...

85
00:04:21,510 --> 00:04:25,320
Missions for national
defense, homeland security,

86
00:04:25,320 --> 00:04:28,383
intelligence and law enforcement purposes.

87
00:04:29,400 --> 00:04:33,330
And I asked the Government
Accountability Office directly,

88
00:04:33,330 --> 00:04:35,380
this is how I got on my first watch list.

89
00:04:37,320 --> 00:04:38,370
Who makes the decision

90
00:04:38,370 --> 00:04:40,530
as to whether a flight
is sensitive or not?

91
00:04:40,530 --> 00:04:43,740
And they told me the
FAA defers to the agency

92
00:04:43,740 --> 00:04:45,330
to make that designation.

93
00:04:45,330 --> 00:04:48,840
So as you can imagine, with the military

94
00:04:48,840 --> 00:04:51,420
self-exempting itself from transponders,

95
00:04:51,420 --> 00:04:55,740
it leaves very little for
flight trackers in the DC Area.

96
00:04:55,740 --> 00:04:58,950
What we normally get is
in the triangle, medical,

97
00:04:58,950 --> 00:05:03,570
news gathering, state and
county police and park police.

98
00:05:03,570 --> 00:05:05,970
Everything else is a ghost helicopter

99
00:05:05,970 --> 00:05:08,490
for the purposes of this discussion.

100
00:05:08,490 --> 00:05:13,490
And so our flight data pyramid
emerges with ADS-B at the top

101
00:05:14,010 --> 00:05:16,170
with the highest data quality,

102
00:05:16,170 --> 00:05:19,800
and only in the DC Area is
it true that this provides

103
00:05:19,800 --> 00:05:22,110
the lowest data quantity

104
00:05:22,110 --> 00:05:24,033
when we're talking about helicopters.

105
00:05:24,990 --> 00:05:26,670
And it didn't feel right having a pyramid

106
00:05:26,670 --> 00:05:28,470
about government
surveillance without this.

107
00:05:28,470 --> 00:05:29,320
So you have that.

108
00:05:30,360 --> 00:05:33,690
Now, we could start to chip
away at this pretty easily

109
00:05:33,690 --> 00:05:36,000
by looking at the
President's daily schedule,

110
00:05:36,000 --> 00:05:37,350
which is public.

111
00:05:37,350 --> 00:05:40,230
And we could scrape this with
beautiful soup or selenium

112
00:05:40,230 --> 00:05:42,540
but they let you download a CSV,

113
00:05:42,540 --> 00:05:45,480
so we just do that every
morning at 8:00 AM,

114
00:05:45,480 --> 00:05:48,840
parse for relevant terms
like Joint Base Andrews

115
00:05:48,840 --> 00:05:50,880
where the helicopter flies to,

116
00:05:50,880 --> 00:05:53,130
and then convert those into Chron jobs

117
00:05:53,130 --> 00:05:55,050
that tweet later in the day,

118
00:05:55,050 --> 00:05:58,260
and voila you have automatic
presidential arrivals

119
00:05:58,260 --> 00:06:00,870
and departures, unless
they take the motorcade,

120
00:06:00,870 --> 00:06:02,580
which does happen.

121
00:06:02,580 --> 00:06:06,000
Now, this type of tracking
what's publicly announced

122
00:06:06,000 --> 00:06:09,510
is also useful for things
like military night

123
00:06:09,510 --> 00:06:11,223
at the local ballpark.

124
00:06:12,420 --> 00:06:14,310
And when you're surveilling the military,

125
00:06:14,310 --> 00:06:15,600
and before we go any further,

126
00:06:15,600 --> 00:06:18,180
it's good to set some ground rules.

127
00:06:18,180 --> 00:06:20,640
In researching this I relied on a talk

128
00:06:20,640 --> 00:06:23,910
by the National Geospatial Programs Office

129
00:06:23,910 --> 00:06:26,760
on how they make the
decision to redact data

130
00:06:26,760 --> 00:06:28,800
from public datasets.

131
00:06:28,800 --> 00:06:30,420
So they consider first,

132
00:06:30,420 --> 00:06:33,750
is this data potentially
useful to an adversary?

133
00:06:33,750 --> 00:06:37,980
And second, is this data unique
before weighing the risks

134
00:06:37,980 --> 00:06:41,730
and benefits of redacting that
information from the public?

135
00:06:41,730 --> 00:06:44,850
They note efforts to
safeguard useful information

136
00:06:44,850 --> 00:06:47,340
that is readily available
through open sources

137
00:06:47,340 --> 00:06:51,123
or observation are unlikely
to reduce vulnerabilities.

138
00:06:52,110 --> 00:06:55,410
On the other side, we have basic
tenants of good journalism.

139
00:06:55,410 --> 00:06:59,250
I try not to do things
automatically, ADS-B bot aside,

140
00:06:59,250 --> 00:07:02,670
provide proof of user specific
interest whenever possible

141
00:07:02,670 --> 00:07:05,700
and thus establishing the public interest,

142
00:07:05,700 --> 00:07:07,020
and source good data

143
00:07:07,020 --> 00:07:10,680
and link to primary
sources whenever possible.

144
00:07:10,680 --> 00:07:14,040
A 1997 court case defined
what a journalist is

145
00:07:14,040 --> 00:07:15,870
for the purposes of journalistic shield.

146
00:07:15,870 --> 00:07:17,610
And our use case is a little different

147
00:07:17,610 --> 00:07:19,650
but I find it informative.

148
00:07:19,650 --> 00:07:22,110
So they say a journalist is engaged

149
00:07:22,110 --> 00:07:26,610
with the intent to use material
sought, gathered or received

150
00:07:26,610 --> 00:07:29,430
to disseminate information to the public

151
00:07:29,430 --> 00:07:32,520
and that such intent
existed at the inception

152
00:07:32,520 --> 00:07:34,473
of the news gathering process.

153
00:07:35,340 --> 00:07:36,600
All of this is to say,

154
00:07:36,600 --> 00:07:39,510
as long as there's no
smokey backroom discussion

155
00:07:39,510 --> 00:07:41,460
where we're trading helicopter secrets

156
00:07:41,460 --> 00:07:45,150
with no intent of surfacing
this information to the public,

157
00:07:45,150 --> 00:07:47,160
I should be within my
First Amendment rights

158
00:07:47,160 --> 00:07:48,843
for the duration of this talk.

159
00:07:50,250 --> 00:07:52,860
And I do a lot of reporting on helicopters

160
00:07:52,860 --> 00:07:54,930
and often the military.

161
00:07:54,930 --> 00:07:59,930
These are two of the
most infamous, famous,

162
00:08:00,840 --> 00:08:03,420
they're the most popular
operators in the DC Area.

163
00:08:03,420 --> 00:08:05,790
The Air Force 1st Helicopter Squadron

164
00:08:05,790 --> 00:08:08,670
responsible for VIP evacuations,

165
00:08:08,670 --> 00:08:11,250
but well known for flying very low

166
00:08:11,250 --> 00:08:15,180
and making emergency
landings on local playgrounds

167
00:08:15,180 --> 00:08:17,343
in these Vietnam era Hues.

168
00:08:18,180 --> 00:08:22,110
And on the right we have the
Army 12th Aviation Battalion

169
00:08:22,110 --> 00:08:23,790
who's responsible for transporting

170
00:08:23,790 --> 00:08:26,880
three and four star generals
to and from the Pentagon,

171
00:08:26,880 --> 00:08:29,640
as well as the
aforementioned night flights.

172
00:08:29,640 --> 00:08:32,310
Now, I wanna try and
exercise at this point,

173
00:08:32,310 --> 00:08:36,300
the event on the left was a
diversity recruiting event.

174
00:08:36,300 --> 00:08:39,000
And we recently saw the
operator on the right

175
00:08:39,000 --> 00:08:43,440
doing an honorary military
flyover at Arlington Cemetery.

176
00:08:43,440 --> 00:08:44,940
I'd like you to raise your hand

177
00:08:44,940 --> 00:08:48,420
if you believe either of those
would require an exemption

178
00:08:48,420 --> 00:08:51,453
from transponder use
as a sensitive flight.

179
00:08:54,030 --> 00:08:56,910
Oh my God, well, okay,
that's what I thought too,

180
00:08:56,910 --> 00:09:00,300
but no transponder on either of those.

181
00:09:00,300 --> 00:09:03,780
So it would be easy to
look at what we're doing

182
00:09:03,780 --> 00:09:05,467
so far on Twitter and say,

183
00:09:05,467 --> 00:09:07,560
"Well, you have an enormous data problem

184
00:09:07,560 --> 00:09:09,810
if you can't see these aircraft."

185
00:09:09,810 --> 00:09:11,310
But I wanted to approach this differently.

186
00:09:11,310 --> 00:09:14,490
I thought if we have hundreds
or thousands of people

187
00:09:14,490 --> 00:09:16,350
looking at our Twitter feed on their phone

188
00:09:16,350 --> 00:09:19,140
that is sensor rich
and high quality camera

189
00:09:19,140 --> 00:09:22,770
and we could just get
them to square their hips

190
00:09:22,770 --> 00:09:24,720
and raise the phone above their head

191
00:09:24,720 --> 00:09:26,340
and crane their neck back.

192
00:09:26,340 --> 00:09:29,250
Well, we could solve the data problem

193
00:09:29,250 --> 00:09:31,600
and it might actually
have health benefits too.

194
00:09:32,970 --> 00:09:35,610
So in October, 2020, we
started CopterSpotter

195
00:09:35,610 --> 00:09:38,070
which is a platform for DC residents

196
00:09:38,070 --> 00:09:41,640
to self-report helicopter
activity over their location.

197
00:09:41,640 --> 00:09:44,220
They do this with a couple pieces of data.

198
00:09:44,220 --> 00:09:47,730
The time, date and geolocation
provided by Twitter

199
00:09:47,730 --> 00:09:52,680
and either a photo, video
or the type of helicopter,

200
00:09:52,680 --> 00:09:53,670
if they know it.

201
00:09:53,670 --> 00:09:57,000
Now, this is a typical
CopterSpotter on the right.

202
00:09:57,000 --> 00:09:58,920
And you all could be
forgiven for thinking that

203
00:09:58,920 --> 00:10:02,130
this is a Blackhawk
helicopter carrying armaments

204
00:10:02,130 --> 00:10:04,320
over a civilian population.

205
00:10:04,320 --> 00:10:07,950
These are in fact auxiliary fuel tanks.

206
00:10:07,950 --> 00:10:11,850
But it is a common concern
amongst DC residents.

207
00:10:11,850 --> 00:10:16,850
And our bot retweets a
spot like this with a map,

208
00:10:17,250 --> 00:10:20,130
a corresponding icon
and probable operators,

209
00:10:20,130 --> 00:10:23,970
in this case for a Blackhawk
helicopter, there are several.

210
00:10:23,970 --> 00:10:26,730
And this creates kind of a rolling feed

211
00:10:26,730 --> 00:10:28,920
of helicopter activity
throughout the area.

212
00:10:28,920 --> 00:10:31,470
And I'm sure you would
all say, that's great,

213
00:10:31,470 --> 00:10:34,110
but how do you get people
to actually do this?

214
00:10:34,110 --> 00:10:39,110
Well, you make it a game and
here's a recent leaderboard.

215
00:10:40,170 --> 00:10:42,930
Our users frequently joke that one day

216
00:10:42,930 --> 00:10:45,300
this could be tradeable as copter coin,

217
00:10:45,300 --> 00:10:47,160
a pseudo currency that can only be earned

218
00:10:47,160 --> 00:10:49,053
by taking photos of helicopters.

219
00:10:49,980 --> 00:10:51,570
That puts it much more
in the tangible realm

220
00:10:51,570 --> 00:10:53,470
compared to a lot of cryptocurrencies.

221
00:10:54,720 --> 00:10:55,650
Quickly, I wanna tell you

222
00:10:55,650 --> 00:10:58,770
how Twitter handles location tags now.

223
00:10:58,770 --> 00:11:00,990
It used to be that you could
tag a specific latitude

224
00:11:00,990 --> 00:11:02,310
and longitude on a map.

225
00:11:02,310 --> 00:11:07,310
And since 2019, they've in favor
used four square locations,

226
00:11:07,560 --> 00:11:10,680
which gives you a bounding
box, by default your city,

227
00:11:10,680 --> 00:11:12,480
which is too large for our purposes.

228
00:11:12,480 --> 00:11:15,540
So we ask users to tag a
specific park or building

229
00:11:15,540 --> 00:11:18,570
and then we make sure that
that is no wider than two miles

230
00:11:18,570 --> 00:11:19,743
in either dimension.

231
00:11:20,640 --> 00:11:23,940
Here is just one of the
few problematic locations

232
00:11:23,940 --> 00:11:27,483
a park where the bounding
box far exceeds the bounds.

233
00:11:28,380 --> 00:11:30,960
And once you have all
of this location data

234
00:11:30,960 --> 00:11:33,120
you can start to solve
some of those problems.

235
00:11:33,120 --> 00:11:36,690
This is the CopterSpotter map
by my colleague, Sam Reese.

236
00:11:36,690 --> 00:11:39,030
And it lets you filter our dataset

237
00:11:39,030 --> 00:11:41,730
by time, date and operator.

238
00:11:41,730 --> 00:11:44,310
So you can find out who
that helicopter was over you

239
00:11:44,310 --> 00:11:47,760
last night, last week or last month,

240
00:11:47,760 --> 00:11:50,220
even if it was the military.

241
00:11:50,220 --> 00:11:53,790
This also lets you submit
directly to our dataset

242
00:11:53,790 --> 00:11:56,040
via a form in case you don't wanna give

243
00:11:56,040 --> 00:11:57,150
your Twitter location

244
00:11:57,150 --> 00:11:59,940
to whoever our Twitter overlord may be

245
00:11:59,940 --> 00:12:02,240
by the time you're all
watching this on video.

246
00:12:03,630 --> 00:12:05,670
So residents really took
heart with this project

247
00:12:05,670 --> 00:12:07,410
over the pandemic.

248
00:12:07,410 --> 00:12:11,220
One of our spotters, Cassie
told us that it made her

249
00:12:11,220 --> 00:12:13,080
feel like she was included in a group

250
00:12:13,080 --> 00:12:15,030
that she didn't even know existed.

251
00:12:15,030 --> 00:12:17,220
And I recently found out how that felt

252
00:12:17,220 --> 00:12:19,803
here at my first DEF
CON, so thank you all.

253
00:12:22,050 --> 00:12:24,540
Don Beyer, one of the
members of the Congressional

254
00:12:24,540 --> 00:12:27,150
Quiet Skies Caucus recognized my efforts.

255
00:12:27,150 --> 00:12:31,620
And we even got a spot from a
guy in a Blackhawk helicopter,

256
00:12:31,620 --> 00:12:33,033
that's pretty cool.

257
00:12:34,110 --> 00:12:38,520
Not everyone was so excited,
but when someone calls you

258
00:12:38,520 --> 00:12:41,520
an anarchist and says the
FBI needs to be watching you

259
00:12:41,520 --> 00:12:43,620
because of what your project is doing,

260
00:12:43,620 --> 00:12:46,785
everyone in this room knows
you're on the right track.

261
00:12:46,785 --> 00:12:51,035
(audience laughing and applauding)

262
00:12:54,720 --> 00:12:59,720
And we collected over
10,000 copter spots in 2021.

263
00:12:59,820 --> 00:13:01,740
At the top you can see the small portion

264
00:13:01,740 --> 00:13:05,010
that would normally be
available with ADS-B data.

265
00:13:05,010 --> 00:13:08,490
The gray area is the unknowns
after we looked at the photos

266
00:13:08,490 --> 00:13:11,160
and tried to correlate
operators to have all of them.

267
00:13:11,160 --> 00:13:15,093
And then 50% of our data
is military operators.

268
00:13:15,990 --> 00:13:18,480
We do acknowledge a
shortcoming with our dataset

269
00:13:18,480 --> 00:13:21,090
which is that they are
voluntarily submitted.

270
00:13:21,090 --> 00:13:24,150
So they do not represent
a measured random sample.

271
00:13:24,150 --> 00:13:26,100
However, you can reasonably expect

272
00:13:26,100 --> 00:13:27,900
that the actual number of events

273
00:13:27,900 --> 00:13:29,913
is higher than what is being reported.

274
00:13:30,810 --> 00:13:34,140
We did notice a drop off
however of over several months,

275
00:13:34,140 --> 00:13:35,640
unique users were going down

276
00:13:35,640 --> 00:13:37,710
and unknown spots were going down as well.

277
00:13:37,710 --> 00:13:40,860
So people were getting better
at identifying helicopters

278
00:13:40,860 --> 00:13:43,650
but we were serving more of
a core group of spotters.

279
00:13:43,650 --> 00:13:48,210
And I wanted this to
engage the whole DC Area

280
00:13:48,210 --> 00:13:49,710
and serve them.

281
00:13:49,710 --> 00:13:53,130
And I looked for ways to use
the existing data to that end.

282
00:13:53,130 --> 00:13:56,070
And of course, once you
have thousands of photos

283
00:13:56,070 --> 00:13:59,550
of helicopters submitted
from the user's perspective,

284
00:13:59,550 --> 00:14:01,350
the next step's pretty obvious,

285
00:14:01,350 --> 00:14:03,240
you create a computer vision program

286
00:14:03,240 --> 00:14:06,420
to automatically identify the helicopters.

287
00:14:06,420 --> 00:14:08,640
These are two V-22 Osprey,

288
00:14:08,640 --> 00:14:11,040
part of the Marine Presidential Squadron

289
00:14:11,040 --> 00:14:13,620
and our computer vision
program identified these

290
00:14:13,620 --> 00:14:17,370
with a 73.9% confidence

291
00:14:17,370 --> 00:14:21,513
when selecting from 19 distinct operators.

292
00:14:22,620 --> 00:14:25,350
If the user also includes
location, we add them

293
00:14:25,350 --> 00:14:27,330
to the dataset and map it as we did

294
00:14:27,330 --> 00:14:30,090
with this Coast Guard MH-65,

295
00:14:30,090 --> 00:14:33,630
part of the Blackjack
Flight Interdiction Program

296
00:14:33,630 --> 00:14:35,763
that keeps small planes out of NCR.

297
00:14:36,930 --> 00:14:38,550
This is done with the aid of Roboflow,

298
00:14:38,550 --> 00:14:39,990
which is a great startup

299
00:14:39,990 --> 00:14:41,970
and has been a great partner for us.

300
00:14:41,970 --> 00:14:44,130
They help beginners and
small teams get started

301
00:14:44,130 --> 00:14:46,890
with computer vision and
they have amazing tools

302
00:14:46,890 --> 00:14:49,920
for collaborative annotation,
and they can even train

303
00:14:49,920 --> 00:14:52,860
and deploy your model nearly code free.

304
00:14:52,860 --> 00:14:56,130
They also have a Roboflow
Universe that is a repository

305
00:14:56,130 --> 00:15:00,120
of open source models that you
can use for your next project

306
00:15:00,120 --> 00:15:01,260
and our data's included.

307
00:15:01,260 --> 00:15:02,820
So I look forward to seeing

308
00:15:02,820 --> 00:15:06,003
all of your helicopter
identification program soon.

309
00:15:07,530 --> 00:15:10,050
So with that, there's one box missing

310
00:15:10,050 --> 00:15:13,350
from our flight data pyramid here,

311
00:15:13,350 --> 00:15:16,020
and it has to do with the fact
that visual identification

312
00:15:16,020 --> 00:15:18,510
is not that usable at night.

313
00:15:18,510 --> 00:15:20,700
And so we looked for
someone who has more data

314
00:15:20,700 --> 00:15:21,660
that we could access.

315
00:15:21,660 --> 00:15:24,570
And we started capturing radio calls.

316
00:15:24,570 --> 00:15:28,290
Now, we thought about
doing this with live ATC,

317
00:15:28,290 --> 00:15:31,830
but turns out separating by
silence is not quite the same

318
00:15:31,830 --> 00:15:32,940
as squelch on a radio

319
00:15:32,940 --> 00:15:34,770
if you're trying to get individual calls.

320
00:15:34,770 --> 00:15:37,560
So instead we deployed a Raspberry Pi

321
00:15:37,560 --> 00:15:40,503
on a balcony near DCA Volunteers.

322
00:15:41,520 --> 00:15:42,805
He knows what we're doing

323
00:15:42,805 --> 00:15:43,638
(Andrew chuckling)

324
00:15:43,638 --> 00:15:48,240
and we use the excellent open
source app RTLSDR-Airband

325
00:15:48,240 --> 00:15:51,000
to capture individual radio calls,

326
00:15:51,000 --> 00:15:56,000
upload them to an FTP server
and tweet them at a sub-account

327
00:15:56,040 --> 00:15:58,773
from a sub-account called UFOs of DC.

328
00:16:00,000 --> 00:16:02,820
They're unidentified, but
you get to identify them.

329
00:16:02,820 --> 00:16:07,470
And someone did, below, noting
the call signs, way points,

330
00:16:07,470 --> 00:16:10,830
routes and zones that were
mentioned in the radio call.

331
00:16:10,830 --> 00:16:13,710
Now the main account picks
this up, builds a map

332
00:16:13,710 --> 00:16:17,160
of the described features
and adds operator information

333
00:16:17,160 --> 00:16:20,920
and then uses FFM peg to
tie the original audio file

334
00:16:22,230 --> 00:16:24,870
to this image to create a video,

335
00:16:24,870 --> 00:16:27,270
in what I think is a really powerful tool

336
00:16:27,270 --> 00:16:29,940
for citizen journalists to create stories

337
00:16:29,940 --> 00:16:31,803
from a difficult to use dataset.

338
00:16:33,000 --> 00:16:34,920
Quickly I wanna show you
guys our Raspberry Pi.

339
00:16:34,920 --> 00:16:36,480
This was a very early version,

340
00:16:36,480 --> 00:16:39,480
we've since upgraded to an Airspy HF+.

341
00:16:39,480 --> 00:16:42,750
And the main thing I learned
is that you cannot trust

342
00:16:42,750 --> 00:16:45,600
the Raspberry Pi's built in WiFi at all.

343
00:16:45,600 --> 00:16:48,870
So instead we use a GLI net router

344
00:16:48,870 --> 00:16:52,020
and the Good Cloud
Interface allows you access

345
00:16:52,020 --> 00:16:54,510
a layer above the Pi and
you can use something

346
00:16:54,510 --> 00:16:57,390
like the Shelly command
line accessible switch,

347
00:16:57,390 --> 00:17:00,870
accessed from the router
to hard reset the Pi

348
00:17:00,870 --> 00:17:02,613
in the event that it crashes.

349
00:17:03,810 --> 00:17:05,610
And with the completion of that box,

350
00:17:05,610 --> 00:17:08,580
we've completed our
flight data pyramid, yay.

351
00:17:08,580 --> 00:17:10,620
Where do we go from here?

352
00:17:10,620 --> 00:17:14,250
Well, I'd like to look at
connecting disparate data sources,

353
00:17:14,250 --> 00:17:17,520
whether that is the DC Police
tweeting an incident report

354
00:17:17,520 --> 00:17:20,880
that includes location and
tying that to ADS-B data

355
00:17:20,880 --> 00:17:23,010
to say "This helicopter's
above you right now

356
00:17:23,010 --> 00:17:25,320
and this is what they're looking for."

357
00:17:25,320 --> 00:17:29,070
Similarly I'd like to look
at news gathering helicopters

358
00:17:29,070 --> 00:17:33,750
and scrape local news sites
just for the trending stories

359
00:17:33,750 --> 00:17:37,320
and identify aerial photos
with computer vision

360
00:17:37,320 --> 00:17:38,970
so that we can say this is likely

361
00:17:38,970 --> 00:17:41,580
what they're over your house for.

362
00:17:41,580 --> 00:17:45,060
We're elevating helicopter
air to air radio

363
00:17:45,060 --> 00:17:46,680
directly to the main feed as well.

364
00:17:46,680 --> 00:17:49,020
This is often two news
gathering helicopters

365
00:17:49,020 --> 00:17:51,390
talking about how they're covering a story

366
00:17:51,390 --> 00:17:55,320
or two police agencies talking
about an active pursuit.

367
00:17:55,320 --> 00:17:57,660
And I'm interested in noise monitoring.

368
00:17:57,660 --> 00:18:02,130
The DCA Airport has a
noise monitoring program,

369
00:18:02,130 --> 00:18:03,750
but it's only along the river,

370
00:18:03,750 --> 00:18:06,390
and it's not publicly accessible via API.

371
00:18:06,390 --> 00:18:08,430
But I'm really curious to ask you guys

372
00:18:08,430 --> 00:18:10,620
how would you apply something like this

373
00:18:10,620 --> 00:18:13,500
to crowdsource a difficult to use dataset

374
00:18:13,500 --> 00:18:16,680
or give citizen journalists
the power to make stories

375
00:18:16,680 --> 00:18:17,643
from that data?

376
00:18:18,480 --> 00:18:20,700
And we have a demonstration this weekend.

377
00:18:20,700 --> 00:18:23,370
If you go to helicoptersofLV on Twitter,

378
00:18:23,370 --> 00:18:25,890
you will see the ADS-B bot in action,

379
00:18:25,890 --> 00:18:28,110
mostly covering sight seeing helicopters

380
00:18:28,110 --> 00:18:29,700
over the Las Vegas region,

381
00:18:29,700 --> 00:18:31,260
but you can tweet your photos at it

382
00:18:31,260 --> 00:18:33,360
for automatic identification.

383
00:18:33,360 --> 00:18:36,750
And I'm gonna hang out at 5:00 PM

384
00:18:36,750 --> 00:18:39,480
right outside Caesars by the sign.

385
00:18:39,480 --> 00:18:42,420
And if you want to come do five
minutes of copter spotting,

386
00:18:42,420 --> 00:18:45,300
say hello, maybe look at
code and give me your ideas,

387
00:18:45,300 --> 00:18:47,070
I would love to hear them.

388
00:18:47,070 --> 00:18:48,762
Thank you all so much.

389
00:18:48,762 --> 00:18:51,353
(audience applauding)