WEBVTT

1
00:00:00.010 --> 00:00:04.050
[Don Roberts] Here at S&T we're developing
a number of tools among them a video

2
00:00:04.050 --> 00:00:08.130
algorithm capable of automatically
detecting bags left behind

3
00:00:08.130 --> 00:00:12.230
and tagging individuals who left those bags.
We're also developing a

4
00:00:12.230 --> 00:00:16.390
suite of video forensic tools that allow
video surveillance systems to work more effectively

5
00:00:16.390 --> 00:00:20.400
and efficiently.

6
00:00:20.400 --> 00:00:24.570
[Robert Sealock] Today we are at New Carollton Metro station. We are testing some of camera

7
00:00:24.570 --> 00:00:28.630
algorithms by placing bags in a variety of locations.

8
00:00:28.630 --> 00:00:32.750
If you have a fairly well performing commercial algorithm
that alerts

9
00:00:32.750 --> 00:00:36.750
maybe only 60 times an hour, once a
minute, when you multiple that by the

10
00:00:36.750 --> 00:00:40.970
50,000 or so video camera in a
mass transit system.

11
00:00:40.970 --> 00:00:45.060
The number of false alerts is quite
overwhelming. Its inconceivable

12
00:00:45.060 --> 00:00:49.310
that even a fully staffed operations center
is going to be able to field that many events over the course

13
00:00:49.310 --> 00:00:53.340
a day. We are looking at a false
alert rate that is appreciably lower

14
00:00:53.340 --> 00:00:57.370
then what's commercially available
and basically baked into the infrastructure.

15
00:00:57.370 --> 00:01:01.420
and the video analytics and the video
management systems that presently exist.

16
00:01:01.420 --> 00:01:05.450
[Shawn Doody] The FOVEA program allows us to

17
00:01:05.450 --> 00:01:09.490
be able to identify a package that's been
left behind and then

18
00:01:09.490 --> 00:01:13.590
to figure out who left it
it behind and then start

19
00:01:13.590 --> 00:01:17.630
to track that person and determine
weather or not this is a

20
00:01:17.630 --> 00:01:22.630
threatful situation and deploy
proper resources to keep the people safe.

21
00:01:22.760 --> 00:01:26.780
[Marianne Deangelus] So we're helping make existing
video surveillance

22
00:01:26.780 --> 00:01:30.870
systems more efficient, more effective, by giving
users tools that help them get through the video faster.

23
00:01:30.870 --> 00:01:34.940
One of the tools within FOVEA is a tool
that we call jump-back and it lets a user

24
00:01:34.940 --> 00:01:39.100
highlight an abandoned object, simply
draw a box around it and jump back

25
00:01:39.100 --> 00:01:43.290
to when that object first appeared.
And from there the user can investigate what are the

26
00:01:43.290 --> 00:01:47.340
circumstances around it. So for instance
it's jumped back to when the

27
00:01:47.340 --> 00:01:51.470
person left the bag. We can actually
begin to bookmark the person in the video

28
00:01:51.470 --> 00:01:55.470
and follow as they move throughout
the station.

29
00:01:55.470 --> 00:01:59.550
So here we're actually piecing together information
from different video cameras.

30
00:01:59.550 --> 00:02:03.570
Now once a user actually followers
a person throughout the station

31
00:02:03.570 --> 00:02:07.600
and understands where they have come
from and where they have gone to

32
00:02:07.600 --> 00:02:11.700
they could then reconstruct that video
and stitch all of those pieces together

33
00:02:11.700 --> 00:02:15.830
into once final video.

34
00:02:15.830 --> 00:02:20.830
[Music]