PERSONAL PROTECTIVE EQUIPMENT DETECTION AND MACHINE POWER CONTROL USING IMAGE PROCESSING

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 603
PERSONAL PROTECTIVE EQUIPMENT DETECTION AND MACHINE
POWER CONTROL USING IMAGE PROCESSING
Bharani T1, Jishin Jayan T2, Vadivel S3
1Bharani T, PG Student, Department of Mechanical Engineering, Bannari amman Institue of Technology, Tamil
Nadu, India
2Jishin Jayan T, Asst. Professor, Department of Mechanical Engineering, Bannari amman Institue of Technology,
Tamil Nadu, India
3Vadivel S, PG Student, Department of Mechanical Engineering, Bannari amman Institue of Technology, Tamil
Nadu, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Safety is defined as the act of protecting oneself
from injury or, in other words, being aware of the presence of
danger. In the new era of technology development and
automation, Image Processing has shown tremendousgrowth
in the 21st century making a better future for society and for
human beings. The incorporationofImageprocessing insafety
is one of the fast growing and promising methods in
promoting safety in industries as well as in protecting human
lives. PPEs are important is any industry as they help to
protect personnel from injury when all other lines of defense
(engineering controls, administrativecontrols) fail. Theaimof
incorporating Image processing in PPE detection is to provide
protection to the person operating machines like cutter or
grinders. Personal Protective Equipment is must for an
operator but usually in the hurry of work some operator
forgot to wear the safety equipment and right away they start
to operate the machine which is completely not a safe process
of working. This may lead or cause anything to the operator.
With the help of image processing technique, the Personal
Protective Equipment is monitored for every operator when
they try to work in the machine, it will not work unless they
wear PPE. The device worksbydetectingProtectiveequipment
with image processing technique and allows access to the
machine by closing the circuit. This helps to rectify the major
danger before occurring itself, proving a better solutiontothe
safety of workers. To sum up PPE detection to avoid accidents
using Image Processing is a greatinitiativetoprovidesafety to
workers in industries.
Key Words: Personal Protective Equipment, safety,
image processing, access to machine.
1.INTRODUCTION
Despite development of science and technology, statistics
from the International Labour Organization (ILO) show that
workplace environments in many countries (e.g., the
European Union) have not improved to the point where the
problem of occupational injuries has been significantly
reduced. As a result, every effort should be made to reduce
the number of accidents or, at the at least, keep the rate
within an acceptable range, which can be achieved through
organizational actions, collective training, or individual
safeguards. Establishing barriers, which plays a key part in
accident prevention, is the traditional strategy to avoiding
loss. Safety barriers are characterized as "physical and/or
non-physical measures meant to deter, control, or mitigate
undesired events or accidents”. There are major
opportunities before it becomes a loss, to prevent or change
an accident sequence of events. The first solution is to alter
the necessary conditions for an event to happen by
eliminating or adjusting the energy characteristics of the
hazard.
1.1 Objective
The objective is to prevent danger before it occurs when
operator operates any machine without proper protective
equipment. With the help of image processing technique, the
Personal Protective Equipment is monitored for every
operator once it is available then only the operator can
perform operations in the machine.
1.2 Methodology
The process for automatically identifying proper PPE use
is outlined, and the steps are as follows:
1. Individuals are discovered, together with their key
point coordinates, in each image captured by an on-site cctv
system using an individual detection model.
2. Using an object detection model, the PPE(s) are
identified and located.
3. Proper PPE identification is accomplished by studying
the geometric correlations between the individual's key
points and the PPE that has been identified.
2. Image Processing and Computer Vision
Current automated PPE compliance monitoring techniques
can be divided into two categories: sensor-basedandvision-
based. The majority of existing vision-basedPPEcompliance
monitoring systems are limited to spotting hard hats,
employed Region-based CNNs (R-CNNs) to detect whether
or not a worker was wearing a hard hat. The image
processing-based protective eyewear detection system
makes use of widely accessible hardware and software

components for ease of use, as well as cloud computing
support for scalability on-demand. Raspberry Pi, Webcam
module, wires, and a relay are among the hardware
components. A Raspberry Pi is a credit card-sized computer
that is primarily used for projects. The webcam module is
used to record video or photos, with the obtained images
being compared to pre-set data to provide a result. Power
supply, camera, and display cables are all linked to the
Raspberry Pi board, which offers a variety of peripherals for
connections. Other electronic devices are connectedtoGPIO
pins. Relays are commonly usedtomanagethepoweraswell
as switch the smaller current values in a control circuit in
control panels, manufacturing, and building automation.
However, because a low voltage is given to the relay coil, a
big voltage can be switched by the contacts, the supply of
amplifying effect can help regulate huge amperes and
voltages.
Fig -1: System Architecture
2.1 Computer Vision
Computer Vision (CV) is the study of automatic information
extraction from images and movies. 3D models, camera
location, object detection and recognition, as well as
categorizing and searching visual material are all examples
of information. CV combines knowledge from a variety of
domains, including image processing, pattern recognition,
mathematics, and AI. One of the project's key goals is to
enable computers to mimic core human vision functionslike
motion perception and scene interpretation. As a result,
visual object tracking hasbeen extensivelyresearched,andit
involves three critical phases in video analysis: detecting
object movement, tracking suchobjectsfromframetoframe,
and analyzing object tracks to recognize their behaviour.
Visual object tracking is essentially based on reliably
estimating the motion status (i.e., location, orientation, size,
etc.) of a target object in each frame of an input image
sequence. Every system is designed to compensate for
human operators' limitations in monitoring a large number
of cameras at the same time. Exploring similar tools and
problems to identify the use of PPE in order to prevent
accidents in the workplace is an intriguing case.
2.2 You Only Look Once (YOLO)
A type of convolutional neural network isa multilayerneural
network known as a convolutional neural network (CNN).
It's a deep learning technique that recognizes and
categorizes photos. It can solve problems like many
parameters and difficult neural network-based training,
resulting in better classification results. An input layer, a
convolutional layer (Conv layer), a transfer functions, a
pooling layer, and a fully connected layer are all common
features of CNNs (FC layer). Local connectionand parameter
sharing are two key aspects of CNNs, whichlimitthenumber
of factors while enhancing detection efficiency. Object
identification algorithms based on classification, such as R-
CNN as well as other categorizationCNN objectidentification
algorithms, are frequently used. The detecting speed, on the
other hand, is slow and cannot bedonein real time.Although
the SSD algorithm does not have the maximum accuracy, it
becomes much fast and equivalent to the YOLO algorithm in
terms of detection speed, and its accuracycanbehigherthan
the YOLO algorithm when the input image sizes are lower.
Whereas the Faster R-CNN algorithm produces more
accurate estimates, it is significantly slower, taking at least
100 milliseconds per image. As a result, the SSD algorithm
was used in the study because of the real-time detection
requirements.
2.3 Experiment
The computer vision equipped camera is used in the
laboratory fortwo purposes:evaluatinghowsuccessfullythe
camera and model recognize protective eyewear, and
collecting image information for learning the Custom Vision
train model. The image analysis model is trained with
internet photographs of persons wearing safety glasses and
then deployed to the camera using the method described in
the previous section. The model is then tested to see if it can
distinguish between people who are wearing PPE and those
who aren't. The collecting of picture data and tagging of
pictures forlearning thecustomvisionmodelisthefirststage
of development. Three of the co-authors worked as lab
employees for this experiment, recording their photos with
and without safety glasses. VLC media player is used to
capture the camera's live stream, which is then parsed into
images. The Custom Vision model is built using the parsed
images collected from the video feed in the second stage.
Selecting a squareborderareaaroundobjectsinphotographs
and applying tags is howimagingtaggingisdone.TheCustom
Vision model is trained with the tagged photos and the
model'santicipated output,whichiscomputedandrecorded,
after all of the images have now been tagged. The trainingset
is then obtained from the Custom Vision tool and attached to
the camera's cloud-baseddigitaltwin,whichisthendeployed
to the on-premises camera.Finally,thenewmodelistestedto
see if it can accurately detect persons, faces, and PPEs. The
next section contains the results of the performance.

Fig -2: Sample recorded image with Mask (PPE)
The precision and recall of the experimental models were
assessed. Precision refers to the percentage of correctly
detected classes. For example, if the model identified 100
photos as dogs and 99 of them were indeed dogs, the
precision would be 100%. The percentage of actual classes
accurately identified is referred to as recall. If there were
100 photos of apples and the model correctly classified80 of
them as apples, the retention would've been 80%. Finally,
mean Average Precision (mAP) represents the overall
performance of the detector across all tags. When using a
high probability threshold to interpret prediction calls, the
system tends to produce findings with high accuracy at the
cost of recall—the detected categories are correct,butmany
are missed. A small chance threshold had the reverse effect:
most of the true classifications were discovered, but there
were more false positives. With all that in mind, the
likelihood of establishing whether or not the person is
wearing PPE before starting work is increased.
3. Result
The model draws a bounding box around each identified
object from a list of predefined classes and generates an
expected output in a controlled environment, but when we
tested this model in a real-world setting, it failed to
recognize safety mask. This model's performancewasweak,
and it was completely inappropriate for use in a safety
system. This base model has considerable problems
detecting various types of PPEs in general, and is unable to
detect safety helmets in particular.
4. CONCLUSIONS
Using object detection with YOLO, this paper proposed a
method for automatically identifying PPE consumption in a
controlled environment. This method achieves a decent
balance between speed and confidence by exploiting YOLO,
which runs in real time and uses relatively few computer
resources. Furthermore, the model can be adjusted to
different scenarios based on specific requirements. Because
the detection is fully automated and does not require
continual human attention, this could have a positiveimpact
on safety engineering. Our ongoing study is to enhance this
model so that it may be used in a broader range of
circumstances. For example, YOLO may be trained to
recognize other types of PPEs so that it can be used to track
the use of many PPEs at the sametime.Additionally,thealert
script might be improved, allowing this method to cover a
broader range of cases. Future studies will be focused on
defining these operatinglimitsandinvestigatingappropriate
applications, including using genuine surveillance films as
input, detecting the use of PPE in a realistic setting, averting
accidents, and improving industry safety monitoring
systems.
REFERENCES
[1] Allen, M.W., Coopman, S.J., Hart, J.L. and Walker, K.L.
(2015) ‘Workplace Surveillance and Managing Privacy
Boundaries’, Labor History, 51(1), pp. 172–200.
[2] Barnich, O. and Van Droogenbroeck, M. (2011) ‘ViBe: A
universal background subtraction algorithm for video
sequences’, IEEE Transactions on Image Processing,
20(6), pp. 1709–1724. doi: 10.1109/TIP.2010.2101613.
[3] Cavazza, N. and Serpe, A. (2009)‘Effectsofsafetyclimate
on safety norm violations: exploring the mediating role
of attitudinal ambivalence toward personal protective
equipment’, Journal of Safety Research, 40(4), pp. 277–
283. doi: 10.1016/j.jsr.2009.06.002.
[4] Egmont-Petersen, M., de Ridder, D. and Handels, H.
(2002) ‘Image processing with neural networks—a
review’, Pattern Recognition, 35(10), pp. 2279–2301.
doi: 10.1016/S0031–3203(01)00178–9.
[5] Enzweiler, M. and Gavrila, D.M. (2009) ‘Monocular
pedestrian detection: Survey and experiments’, in IEEE
Transactions on Pattern Analysis and Machine
Intelligence, pp. 2179–2195. doi:
10.1109/TPAMI.2008.260.
[6] Guo, X., Chen, L. and Shen, C. (2016) ‘Hierarchical
adaptive deep convolution neural network and its
application to bearing fault diagnosis’, Measurement:
Journal of the International Measurement
Confederation, 93, pp. 490–502. doi:
10.1016/j.measurement.2016.07.054.
[7] Haritaoglu, I., Harwood, D. and Davis, L.S. (2000) ‘W4:
Real-time surveillance of people and their activities’,
IEEE Transactions on Pattern Analysis and Machine
Intelligence, 22(8), pp. 809–830. doi:
10.1109/34.868683.
[8] Krizhevsky, A., Sutskever, I. and Hinton, G.E. (2012)
‘ImageNetClassificationwithDeepConvolutional Neural
Networks’, Advances In Neural Information Processing
Systems, pp. 1–9.
doi:http://dx.doi.org/10.1016/j.protcy.2014.09.007.
[9] Newman, J.A., Beusenberg, M.C., Shewchenko, N.,
Withnall, C. and Fournier, E. (2005) ‘Verification of
biomechanical methods employed in a comprehensive

study of mild traumatic brain injury and the
effectiveness of American football helmets’, Journal of
Biomechanics, 38(7), pp. 1469–1481. doi: 10.1016/j.
jbiomech.2004.06.025.
[10] Redmon, J. and Farhadi, A. (2017) ‘YOLO9000: Better,
Faster, Stronger’, Conference on Computer Vision and
Pattern Recognition, 7(3). doi:
10.1142/9789812771728_0012.
[11] Valera, M. and Velastin, S.A. (2005) ‘Intelligent
distributed surveillance systems: a review’, IEE
Proceedings—Vision, Image, and Signal Processing,
152(2), p. 192. doi: 10.1049/ip-vis:20041147.

PERSONAL PROTECTIVE EQUIPMENT DETECTION AND MACHINE POWER CONTROL USING IMAGE PROCESSING

More Related Content

Similar to PERSONAL PROTECTIVE EQUIPMENT DETECTION AND MACHINE POWER CONTROL USING IMAGE PROCESSING (20)

More from IRJET Journal (20)

Recently uploaded (20)

PERSONAL PROTECTIVE EQUIPMENT DETECTION AND MACHINE POWER CONTROL USING IMAGE PROCESSING