IEEE Consumer Electronics Magazine - July/August 2022 - 22

A Design of Smart UVM for New Retail
recognition processing brought by the binocular camera. A new neural network
structure-the YOLOv3-TinyE is proposed based on YOLOv3-tiny model. Based on the
dataset of 21,000 images captured in real scenarios containing 20 different type of
beverages, the comparison experimental results show that YOLOv3-TinyE model
achieves the mean average precision of 99.15%, and the inference speed is 2.91 times
faster than that of YOLOv3 model, and the detection accuracy of YOLOv3-TinyE model
based on binocular vision is higher than that based on monocular vision. The results
suggest that the designed method achieves the goal in terms of inference speed and
average precision, that is, it is able to satisfy the requirements for real-world
applications.
& NOWADAYS, THE INCREASE of the cost for labor
and rental of the retail shop drives the traditional
retail industry to change urgently. With
the development of the Internet of Things (IoT),
big data, artificial intelligence (AI), and sensor
technology, as well as the popularity of mobile
payment, unmanned retail mode, and the " new
retail " concept are proposed,1,2 which are revolutionizing
the consumer shopping experience,
and in the meantime, profiting both the consumers
and the retailers. Applying these emerging
technologies to revamp brick-and-mortar stores
presents an important paradigm for the realization
of new retail. Companies, such as Amazon,
Alibaba group, and JD position themselves as
players in the industry. Amazon is proud of its
automated Go grocery store concept based on
computer vision in 2016,3 which is able to record
the items picked up by the consumers, and automatically
detect when products are taken from
or returned to shelves. Besides, it also keeps
track of items in a virtual cart. Alibaba has
launched an experimental cashier-less store
called " Tao Cafe " to explore new modes of brick
and mortar retail combined with e-commerce.4
As a part of Alibaba's New Retail strategy, the
Hema stores bring together the best of offline
and online shopping.5 JD.ID X-mart leverages
cutting-edge technologies, such as radio frequency
identification (RFID), facial recognition,
and image recognition, to track retail activity.
In the new retail mode featured by unstaffed
retail shop, unmanned vending machine (UVM) is
the key instrument and a formof unmanned store,
there are even emerging " micro-supermarkets "
based on smart UVMs. The traditional UVM
shows some problems, such as high loss rate and
inconvenient management of goods. In contrast,
the smart UVM that integrated various sensors is
not only a container of goods, but also available
for item access recognition, goods management,
and automatic payment. As a good solution,
smart UVMs based on RFID technology6 attach
RFID tags on goods to realize nontouch automatic
identification, thereby solving the problem of
multipoint shopping settlement and goods management.7
However, as the disadvantage, the recognition
effect of metal and liquid is poor and
with too high operation cost. SomeUVMs achieve
goods identification and self-service payment by
scanning QR code of the customers,8 however,
such an open retail environment is highly
exposed to security risks due to the low technical
threshold. Similarly, although the UVMs can
detect the items being moved based on the gravity
sensing method, it cannot sense what the customer
are taking.
With the development of artificial intelligence
technology, the technologies adopted by smart
UVMs include but not limited to access control by
biological recognition, and machine vision techniques,
such as object detection and classification.9,10
In 2018, an intelligent container equipped
with AI technologies is introduced by Tencent
Youtu Laboratory at the Computer Vision Summit,11
which integrates various technologies,
including deep learning techniques, and visual
product identification algorithms. To improve the
item identification rate based on image recognition,
Megvii Technology proposes a large-scale
retail product checkout dataset to address the
problem, that is, the lack of a high-quality images
dataset of the products to purchase.12 In,13
benchmark dataset for smart unmanned vending
a
22
IEEE Consumer Electronics Magazine
http://www.JD.ID
IEEE Consumer Electronics Magazine - July/August 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - July/August 2022
