WEBVTT 1 00:00:00.328 --> 00:00:02.911 (gentle music) 2 00:00:11.310 --> 00:00:13.020 Safeguarding the American public 3 00:00:13.020 --> 00:00:14.960 requires constant vigilance 4 00:00:14.960 --> 00:00:18.020 and video cameras can help by identifying threats, 5 00:00:18.020 --> 00:00:20.713 but they also raise concerns about privacy. 6 00:00:22.300 --> 00:00:24.770 However, security and privacy 7 00:00:24.770 --> 00:00:27.223 need not be mutually exclusive. 8 00:00:28.900 --> 00:00:30.700 To address privacy concerns 9 00:00:30.700 --> 00:00:32.980 while maintaining high levels of protection, 10 00:00:32.980 --> 00:00:36.480 the Department of Homeland Security Science and Technology 11 00:00:36.480 --> 00:00:38.510 or DHS S&T, 12 00:00:38.510 --> 00:00:41.823 created the Privacy Technology Demonstration. 13 00:00:43.010 --> 00:00:44.520 For the first time ever, 14 00:00:44.520 --> 00:00:48.380 DHS S&T initiated an industry challenge 15 00:00:48.380 --> 00:00:50.740 focused on privacy-enhancing technologies 16 00:00:50.740 --> 00:00:53.760 that protect an individual's anonymity on video 17 00:00:53.760 --> 00:00:57.433 by digitally altering and obscuring their facial images. 18 00:00:59.040 --> 00:01:00.520 Any company with technology 19 00:01:00.520 --> 00:01:02.860 that can conceal human faces in video 20 00:01:02.860 --> 00:01:06.073 could apply to have their tools tested in the challenge. 21 00:01:07.090 --> 00:01:09.420 Industry response was enthusiastic, 22 00:01:09.420 --> 00:01:12.230 as S&T received tremendous interest 23 00:01:12.230 --> 00:01:14.833 from technology companies around the world. 24 00:01:16.310 --> 00:01:18.370 With specific criteria in mind, 25 00:01:18.370 --> 00:01:19.900 five companies were selected 26 00:01:19.900 --> 00:01:21.570 from the larger group of applicants 27 00:01:21.570 --> 00:01:24.640 to participate in the industry challenge. 28 00:01:24.640 --> 00:01:27.730 Each of the selected companies known as performers, 29 00:01:27.730 --> 00:01:28.563 then entered 30 00:01:28.563 --> 00:01:31.350 into a cooperative research and development agreement 31 00:01:31.350 --> 00:01:33.210 with S&T. 32 00:01:33.210 --> 00:01:35.060 In February of 2021, 33 00:01:35.060 --> 00:01:37.030 the performers were given an opportunity 34 00:01:37.030 --> 00:01:39.023 to demonstrate their technology. 35 00:01:40.870 --> 00:01:42.280 Using video recordings 36 00:01:42.280 --> 00:01:44.540 shot with a diverse group of volunteers, 37 00:01:44.540 --> 00:01:45.920 S&T created 38 00:01:45.920 --> 00:01:49.350 a series of mock security footage challenge videos, 39 00:01:49.350 --> 00:01:52.680 and the performers set out to find and hide the faces 40 00:01:52.680 --> 00:01:54.660 of all the participants. 41 00:01:54.660 --> 00:01:56.960 These videos realistically simulated 42 00:01:56.960 --> 00:02:00.103 people going through checkpoints and other security areas. 43 00:02:01.970 --> 00:02:03.300 The challenge was to see 44 00:02:03.300 --> 00:02:05.940 how well the privacy-enhancing technologies 45 00:02:05.940 --> 00:02:09.920 of each of the performers worked in real world scenarios. 46 00:02:09.920 --> 00:02:11.140 Their technology suites 47 00:02:11.140 --> 00:02:14.500 had to find and hide every face in each of the videos 48 00:02:14.500 --> 00:02:17.230 without obscuring any other objects or images 49 00:02:17.230 --> 00:02:19.320 that were in the video frames. 50 00:02:19.320 --> 00:02:22.610 Although the tool sets generally worked as intended, 51 00:02:22.610 --> 00:02:25.030 performance varied across different systems 52 00:02:25.030 --> 00:02:29.140 as hiding faces in some video sections proved difficult. 53 00:02:29.140 --> 00:02:32.440 For instance, some technologies did not detect and hide 54 00:02:32.440 --> 00:02:35.960 partially out-of-frame or partly clipped faces, 55 00:02:35.960 --> 00:02:39.033 while other tools were disrupted by rapid motion, 56 00:02:41.030 --> 00:02:43.993 and crowded conditions post a challenge for others. 57 00:02:46.460 --> 00:02:49.280 Some performers had difficulty finding faces 58 00:02:49.280 --> 00:02:50.563 at different angles, 59 00:02:54.870 --> 00:02:58.433 while other systems had challenges finding smaller faces. 60 00:03:02.180 --> 00:03:03.030 Interestingly, 61 00:03:03.030 --> 00:03:05.480 a number of tools were unable to differentiate 62 00:03:05.480 --> 00:03:07.210 between real human faces 63 00:03:07.210 --> 00:03:10.590 and faces printed on t-shirts or other items. 64 00:03:10.590 --> 00:03:14.140 Meanwhile, some systems erroneously hid areas of the video 65 00:03:14.140 --> 00:03:16.780 that did not have any faces at all. 66 00:03:16.780 --> 00:03:19.910 Over the course of the Privacy Technology Demonstration, 67 00:03:19.910 --> 00:03:23.610 S&T validated that there are some promising technologies 68 00:03:23.610 --> 00:03:26.120 that could enable situational awareness 69 00:03:26.120 --> 00:03:28.030 while still safeguarding privacy 70 00:03:28.030 --> 00:03:32.190 and deterring unauthorized face recognition surveillance. 71 00:03:32.190 --> 00:03:34.970 S&T also identified specific areas 72 00:03:34.970 --> 00:03:36.800 where industry needs to improve, 73 00:03:36.800 --> 00:03:38.680 including the handling of environments 74 00:03:38.680 --> 00:03:41.750 with large groups of people and rapid movement. 75 00:03:41.750 --> 00:03:44.950 Going forward, S&T plans to continue engaging 76 00:03:44.950 --> 00:03:47.160 with industry and other interested parties 77 00:03:47.160 --> 00:03:49.320 to measure and compare the performance 78 00:03:49.320 --> 00:03:51.220 of privacy-enhancing technologies 79 00:03:51.220 --> 00:03:54.320 line:15% for a safer and more secure nation. 80 00:03:54.320 --> 00:03:58.550 line:15% To learn more about S&T's Privacy Technology Demonstration 81 00:03:58.550 --> 00:04:00.690 line:15% and other related security measures, 82 00:04:00.690 --> 00:04:04.657 line:15% direct inquiries to peoplescreening@hq.dhs.gov.