Page 175 - Kaleidoscope Academic Conference Proceedings 2020

AI-BASED W-BAND SUSPICIOUS OBJECT DETECTION SYSTEM FOR MOVING
PERSONS USING GAN: SOLUTIONS, PERFORMANCE EVALUATION AND
STANDARDIZATION ACTIVITIES

1
1
2
1
1
Yutaka Katsuyama ; Keping Yu ; San Hlaing Myint ; Toshio Sato ; Zheng Wen ; Xin Qi 1
1
Global Information and Telecommunication Institute, Waseda University, Japan
2
School of Fundamental Science and Engineering, Waseda University, Japan
ABSTRACT various suspicious object detection systems to increase the
probability of suspicious object recognition. Therefore, it is
With the intensification of conflicts in different regions, the very necessary to study AI-based W-band suspicious object
W-band suspicious object detection system is an essential detection systems for moving persons [6].
security means to prevent terrorist attacks and is widely used
With the expansive development of AI technologies, we will
in many crucial places such as airports. Because artificial
usher in the fourth industrial revolution led by AI. It makes
intelligence can perform highly reliable and accurate services
our lives faster, easier, and safer. Cooperating with image
in the field of image recognition, it is used in suspicious
recognition technologies [7], [8], AI can help suspicious
object detection systems to increase the recognition rate for
object systems increase the recognition probability. To this
suspicious objects. However, it is challenging to establish
end, it is necessary to build a suspicious object database
a complete suspicious object database, and obtaining
by means of simulation and the collection of images from
sufficient millimeter-wave images of suspicious objects from
active/passive imagers to assist in AI-based recognition
experiments for AI training is not realistic. To address
technologies. Compared with various mature AI technologies
this issue, this paper verifies the feasibility to generate a
for image recognition, it is difficult to establish a suspicious
large number of millimeter-wave images for AI training by object database. However, the suspicious object database is
generative adversarial networks. Moreover, we also evaluate very important because it directly affects the recognition rate
the factors that affect the AI recognition rate when the of suspicious objects by AI technologies.
original images used for CNN training are insufficient and
Image generation via a generative adversarial network (GAN)
how to increase the service quality of AI-based W-band
[9] is a new idea to build a suspicious object database.
suspicious object detection systems for moving persons. In
GAN is a deep-learning model, which is regarded as one
parallel, all the international standardization organizations
of the most promising methods for unsupervised learning
have been collectively advancing the novel technologies of
on complex distributions in recent years. In this suspicious
AI. We update the reader with information about AI research
object detection system, we try to supplement the suspicious
and standardization related activities in this paper.
object database by generating some images through the GAN.
The images in the suspicious object database are generated
Keywords - Artificial intelligence, generative adversarial
in three ways: simulation, collection of images from
network, millimeter-wave imaging, suspicious object
active/passive imagers, and GAN. For this suspicious object
detection system
detection system, different types and number of images are
used for training with a fixed AI model, the recognition rates
1. INTRODUCTION
for different suspicious objects are different. This paper
With the intensification of conflicts in various regions, will evaluate the suspicious object database for an AI-based
terrorist attacks occur now and then. Many innocent people W-band suspicious object detection system to get a higher
have lost their lives due to terrorist attacks and it has recognition rate for suspicious objects.
threatened our daily lives and property. How to prevent AI technologies have recently attracted much attention from
terrorist attacks has become a hot topic that gets a great the research and standardization communities. National
concern worldwide. The suspicious object detection system and multi-national funded research projects have progressed
[1] is an effective way to prevent the occurrence of terrorist worldwide. International Telecommunication Union -
attacks which have already largely taken place in public places Telecommunication Standardization Sector (ITU-T) started
such as airports, railway stations, etc [2], [3]. However, it AI standardization activities in 2016. In March 2017,
is not wise to perform security checks one by one at the ITU-T Q5/SG20 initiated a new draft supplement ITU-T
entrance of each gate since this will result in bottlenecks and Y.Sup.AI4IoT (ex TR.AI4ToT; Y. AI4SC) to examine how
crowding. It is therefore necessary to perform automated AI could step in as the saviour and bolster the intent of
suspicious object detection on moving people. In addition, urban stakeholders to deploy IoT technologies and eventually
artificial intelligence (AI) [4] and millimeter-wave (MMW) transition to smart cities. Moreover, many AI-related ITU-T
imaging [5], as emerging technologies, have been applied to Focus Groups are now widely used to address industry

978-92-61-31391-3/CFP2068P @ ITU 2020 – 117 – Kaleidoscope