Page 176 - Kaleidoscope Academic Conference Proceedings 2020

2020 ITU Kaleidoscope Academic Conference

needs as they emerge, and when they are not covered achieve this goal, we build a suspicious object database for
within an existing Study Group. These include, but not CNN training by the simulation and experiments. However,
limited to, ITU-T Focus Group on AI for autonomous it is challenging to obtain sufficient millimeter-wave images
and assisted driving (FG-AI4AD), ITU-T Focus Group on of suspicious objects from experiments. In this paper, we
"Environmental Efficiency for Artificial Intelligence and try to use the GAN to generate the millimeter-wave images
other Emerging Technologies" (FG-AI4EE), ITU-T Focus for CNN training and evaluate its feasibility. In addition to
Group on "Artificial Intelligence for Health" (FG-AI4H), this, we also analyze the factors that affect the AI recognition
and ITU-T Focus Group on Machine Learning for Future rate in this system. Finally, we summary the standardization
Networks including 5G (FG-ML5G). related activities for AI technologies to update the readers the
The contributions of this paper are listed as follows. latest information for the fourth industrial revolution led by
1) We propose a solution to an AI-based W-band suspicious AI.
object detection system for moving persons. Compared with
the traditional solution, it can provide non-stop automatic In order to improve the safety of public places, especially in
monitoring with W-band unidentified object detection for densely populated areas, the safety inspection for suspicious
densely populated places. Moreover, AI technologies are objects should be performed automatically and efficiently.
employed to increase the recognition rate of suspicious However, it is unwise to conduct security checks one by
objects in this system. one at each entrance, as this will cause people crowded.
2) In this case, we evaluate the factors that affect the It is necessary to perform suspicious object detection on
recognition rate of suspicious objects and analyze how to the moving people automatically. As illustrated in Figure
increase the service quality of AI-based W-band suspicious 1, the target of this paper is to develop sensing / imaging
object detection systems for moving persons. with AI-based W band (75-110GHz) [10] technologies to
3) We describe the recent progress in the standardization recognize suspicious objects on moving persons.
of AI component technologies in ITU-T and other
standards-oriented organizations.
2.2 System architecture
The rest of this paper is organized as follow. In Section 2, we
introduce an AI-based W-band suspicious object detection システム統合化技術の構成
system for moving persons. More details of suspicious object Primary screening Secondary screening
databases to support AI-based recognition technologies are Location Ａ Location B
presented in Section 3. The image generation via GAN is Primary screening Secondary
screening
given in this section. We evaluate the performance of GAN Active radar [Primary screening]
used for CNN training when original images are insufficient Suspicious person detection Sensor fusion/suspicious
object detection system
system Passive
in Section 4. In Section 5, we introduce the standardization Visible light camera or infrared camera imager Suspicious Suspicious object
Hybrid person database
Active
of AI. Finally, the paper is concluded in Section 6. Location information acquisition imager imager detection
system
Suspicious person detection Hybrid Signal AI-based
Active radar imager proces suspicious object
Notification to staff [Secondary screening] sing recognition
2. AI-BASED W-BAND SUSPICIOUS OBJECT Suspicious Person/ Camera image Imager Imager Suspicious object
object tracking data Suspicious person info raw data raw data recognition result
DETECTION SYSTEM FOR MOVING PERSONS Suspicious Object/Person Network System (2) Person
(1) Suspicious object/person network integrated server tracking server
Network control for
2.1 Objective Collect and record multiple imager data Network interface tracking method for
A suspicious object
suspicious
object/person collection and distribution with imager multiple imagers
information
confidential © 2019 Toshiba Infrastructure & systems Corporation 46
Figure 2 – System architecture
To perform suspicious objects detection without stopping
the flow of people, this system uses a staged screening
method (primary screening/secondary screening) to identify
suspicious objects hidden in the human body. As shown
in Figure 2, in primary screening, the system will use
W-band radars with multiple visible light cameras to detect
suspicious persons 15 meters away. If a suspicious person
Figure 1 – The objective of designed system
is determined, the security staff will guide this person to the
The objective of this paper is to propose a solution, evaluate secondary screening. In the secondary screening, the system
the performance, and update the standardization status on will develop W-band hybrid imagers combined with visible
an AI-based W-band suspicious object detection system light cameras to detect suspicious objects within 5 meters.
for moving persons. With the W-band unidentified object In this two-stage screening process, the sensing/imaging
detection technologies, we provide a non-stop suspicious information (millimeter-wave images) is paired with existing
object detection system for moving persons, which can be visible light camera images, so that security staff can
employed for densely populated places such as airports. In quickly locate the suspicious person and request further
parallel, AI technologies are applied to this system to increase security checks. Moreover, the system will support tracking
the recognition rate for suspicious objects in this solution. To suspicious persons when moving in different regions.

– 118 –