Instrumentation & Measurement Magazine 25-2 - 27

Deep Learning-Based Robot
Vision: High-End Tools for Smart
Manufacturing
Hui Zhang, Li Zhu Liu, He Xie, Yiming Jiang, Jian Zhou, and Yaonan Wang
W
ith the development of smart manufacturing
technology, manufacturers all over the world
have begun using automated processing and
unmanned factories. China already has many high-end production
lines that use intelligent robots, where Artificial
Intelligence (AI) and robot vision play a critical role in facilitating
robot-driven production. This paper provides a review
of the state-of-the-art in Deep Learning (DL) techniques and
the applications of DL-based robot vision in smart manufacturing,
with an emphasis on the contributions made by China.
We conclude by summarizing challenges and prospects in the
area.
Background of Smart Manufacturing
The manufacturing industry occupies a central position in
China's national economy and is a foundational pillar of enterprise
in the country. In 2015, the manufacturing sector of
China accounted for roughly 30% of its GDP [1]. Since 2000,
China has dominated the largest share of global manufacturing
output, but the degree of automation in its production
lines remains low. With the trending toward an aging population
and decreasing demographic dividends, complete
automation in manufacturing has emerged as a pressing
issue for China. Smart manufacturing is thought to be an efficient
approach to solve this issue. Smart manufacturing can
help improve the traditional manufacturing industry, promote
high-end technological innovation, reduce the costs of
production, and achieve the digital transformation to smart
factories. The smart robot is essential enabling equipment for
smart manufacturing in the future. The " Made in China 2025 "
plan has noted that research on and the development of innovative
manufacturing equipment, including industrial robots
capable of intelligent decision-making and automatic task
execution, is an important project for accelerating industrial
transformation.
Robot vision is one of the most efficient tools for smart
robots to process data from the world. The development of
robot vision has enhanced the perceptual capabilities of robots
and helped expand the role of industrial robots from
April 2022
single mechanical operations to complex operations, such
as workpiece measurement, defect detection, and machining.
However, the demand for robot vision is rising with
the increasing complexity of manufacturing. For example,
high-quality manufacturing requires robot vision to provide
accurate positional information, complex production environments
require it so that robots are adaptable, and human-robot
collaboration and multi-robot collaboration in manufacturing
require robot vision with capabilities of scene analysis.
These demands are challenging to satisfy for traditional processing
algorithms, which are suitable only for specific scenes.
With the development of AI technology, DL offers a novel solution
to manufacturing problems. It enables the extraction of
multi-level features from images through a complex network
design to significantly improve the accuracy of detection and
comprehension of the scene by the robot. In addition, research
on transfer learning in DL has drastically improved the adaptability
of robots to different scenes in smart manufacturing [2].
Advanced AI vision-based robots that combine robots with
DL are now widely used in smart manufacturing, as shown
in Fig. 1.
Deep Learning-based Robot Vision
System
The robot vision system is the basic platform for the robot to
achieve visual perception. A basic DL-based robot vision system
consists of lighting, a camera, a processing chip, and a
robot control unit. The lighting and camera determine the
quality of the acquired image. DL algorithms are usually deployed
on processing chips and play an essential role in image
processing-for instance, in object segmentation [3], pose estimation
[4], and object detection [5]. The processing chip uses
DL algorithms to extract information on the position and orientation
of the object in the received images and sends it to
the control unit. The control unit adjusts the pose of the robot
in real time based on the extracted visual information. In
addition, to determine the transformation between robot and
camera coordinate systems, kinematics modeling and camera
calibration are necessary.
IEEE Instrumentation & Measurement Magazine
1094-6969/22/$25.00©2022IEEE
27
Instrumentation & Measurement Magazine 25-2

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-2
