Safety Helmet Detection in Engineering and Management
0 likes79 views
This document describes a study that developed a deep learning-based model using the YOLO-V3 algorithm to detect whether workers on a construction site are wearing safety helmets in real-time. Video from the construction site is analyzed by the model, which can identify people and determine if they have a helmet on. If a worker is identified without a helmet, an alert is sent via SMS and buzzer to notify the worker and manager. The model achieved 92% accuracy in helmet detection testing, improving on previous methods. The goal is to help prevent accidents and injuries by monitoring helmet usage and providing alerts when safety protocols are not followed.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1123
Safety Helmet Detection in Engineering and Management
Jyothika R1, Shweta Salapur2, Mrs. Swathi Sridharan 3 Font Size 12
1,2 Student Dept. of Information Science and Engineering, BNM Institute of Technology, Karnataka, India
3Assistant Professor, Dept. of Information Science and Engineering, BNM Institute of Technology, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract— Due to a lack of knowledge about safety
helmets, accidents and injuriesonconstructionsitesarenow
increasingly common. Worker supervision by hand is
challenging and ineffective. This study is visually checking
the construction site to see if anyone is wearing a safety
helmet, and then notifying the worker and manager with a
sound BUZZER and an SMS if they are. In order to recognise
a safety helmet in real time at a building site, we built a deep
learning-based technique. Helmet detection was done, and
the experimental results indicate that less than 90% of
people wore helmets.Comparedtopreviouslyusedmethods,
our model has an accuracy rate of 92%. The YOLO-V3
algorithm, which is based on convolutional neural networks,
is used in the method that is being discussed.
Key Words: Accidents, BUZZER, SMS, YOLO-V3, Deep
Learning
1.INTRODUCTION
Currently, complex structures and industries are expanding
quickly over the globe, requiring a large workforce on
construction sites. Unexpected accidents and injuries are
more likely to occur on a construction site because of the
complicated environment. In order to prevent these
occurrences, safety helmets are required in this region. A
significant amount of unstructured image data is available
on-site thanks to thevideomonitoringequipment.According
to the State Administration of Work Safety's accident
statistics from 2015 to 2019, of the 80 construction
accidents that were reported, 57 occurred as a result of
improper use of safety helmets by the workers, making up
69 percent of the total. number of mishaps on construction
sites, traditional safety helmet wear checking is quite
challenging and requires a lot of manual labor. Additionally,
using a visual monitor forces inspectors to spend a lot of
time staring at the screen, which can be ineffective. On
construction sites, researchers create deep learning-based
ways to identify safety helmet use, which can help prevent
accidents, injuries from false alarms, and reduced accuracy
rates. This study used the real-time object detectionmethod
YOLO-V3 (You Only Look Once). utilises deep convolution
neural network characteristics.Yolo-V3hastheadvantageof
being a lot faster and more accurate than other networks.
The main goal of this study is to use live streaming to
monitor the building site, identify any workers who are not
wearing safety helmets, and notify them via SMSand buzzer.
2. RELATED WORKS
To prevent injuries to workers at the construction site, it is
crucial to keep an eye on the area and be promptly alerted if
someone is not wearing a safety helmet. According to
physiologic research, people are only capable of monitoring
two signals at once with an accuracy rate of less than 70%
[1]. A multidisciplinary field related to computer vision,
pattern recognition, signal processing, communication,
embedded computing, and image sensor is known as
intelligent multi camera video surveillance. The scales and
complexities ofcamera networksaregrowingassurveillance
technologies advance quickly, and the monitored
environments are getting more complex and denser [2]. A
video surveillance application that automatically analyses a
motorbike rider's helmet use shows promise because
helmets are crucial for preventing brain injuries in traffic
accidents. In order to segment the objects, a Gaussian-N
mixing model is used (GMM). In order to recognize
motorcycles in the foreground objects that have been
labelled, the suggested system then adapts a faster region-
based convolutional neural network (faster R- CNN) [3].
Recently, it was recognized that employing a Convolutional
Neural Network (CNN) or Deep Learning, digital image
pattern recognition and feature extraction have been
successful over the years. The effectiveness of utilizing a
convolutional neural network for feature extraction and
pattern detection in digital images. Wearing a helmet before
entering the workplace is a requirementforthefactorysince
the environment of the steel industry workshop is
complicated and there may be unanticipated dangers.
Helmet testing is a crucial component of the intelligent
monitoring system for steel plant staff, since it allowsfor the
monitoring of this condition. Accuracy of quicker RCNN
algorithms decreases in dim light and complex backgrounds
[5]. employing object segmentation and background
subtraction in the surveillancefootage. Usingvisual cuesand
binary classification, it then establishes whether the bike
rider is wearing a helmet or not. Results from the
experiment show that the accuracy for detecting bike riders
and violators is 98.88 percent and 93.80 percent,
respectively. A frame is processed on average in 11.58
milliseconds, which is suitable for real-time use[6]. Thetask
of automated motion detection in traffic monitoring is
difficult. This study develops a technique to get meaningful
data from security cameras for tracking moving objects in
digital films. The outcomes demonstrate that the suggested
method recognises and tracks moving objects in urban](https://image.slidesharecdn.com/irjet-v9i7213-221022093642-411b7583/85/Safety-Helmet-Detection-in-Engineering-and-Management-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1124
movies with success [7]. The majority of motorcycle riders
are seen to drive without wearing helmets. In most cases of
vehicle accidents, wearing a helmet can lower the risk of
head and severe brain injuries to motorcycle riders. It can
only be used to identify helmet wea rings; it cannot be used
to detect triple riding on a motorcycle [8].
3. PROPOSED METHODOLOGY
The project has been implemented by dividing the entire
project into three modules.
They are:
Data Collection
Data Pre-processing
Detection Of Helmet
A. DATA COLLECTION
In this module weare collecting livefacedatasetforface
recognition for every person we are collection 200 images to
train the model. Data collection is the process of gathering
information on specific areas with the a I’m of evaluation. In
this study each trained color image is converted as gray
image.
B. DATA PREPROCESSING
In this module we are applying image resizing and
normalizing the image using OpenCV. And building the
variousLBPHalgorithmforfacerecognition.
Pseudo code:
Procedure Building Model ()
Input: Cleaned Data
Output: Pre-Trained Model
Step 1: Read the dataset using CV2
Step 2: Extract the face features
Step 3: Convert into Numerical Array
Step 4: Build model
Step 5: Train Model using data
Step 6: Save the pre-trained model
C. DETECTION OF HELMET
In this module we are taking input camera and loading the
yolov3 model to identify the people and count the
persons inthe frame then find outthey are wearinghelmet
or not, if they are not wearing helmet system will send a
lert SMS and make BUZZER.
Pseudo code:
Procedure Helmet Detection ()
Input: Camera frame
Output: Alarm if no helmet
Step 1: Read frame
Step 2: Load pre trained model
Step 3: Count person
Step 4: For all the person
Step 5: Detect the helmet
Step 6: If helmet detected continue
Step 7: else
Step 8: Identify the face and send alert sms
Step 9: Return result
D.ALGORITHM
The classifier YOLO-V3 In contrast to YOLOv2, the pooling
layer is cancelled in the YOLOv3 network, which only has a
convolutional layer. The size of the output feature map can
be altered by varying the convolutional layer's stepsize.The
YOLOv3 network employs the DARKNET53 feature
extraction network. YOLOv3 employs a tinyfeaturemapand
incorporates the concept of a "pyramid feature map" (FPN).
The usage of large feature maps allows for the detection of
large objects. The rationale for this strategy is that the
higher-level network features include richer semantic
information while the lower-level network features have
more accurate location characteristics. The YOLOv3
prediction layer is able to precisely find the object and
correctly classify itbecausetothefeaturemap'scombination
of accurate location information from low-level network
feature maps and rich semantic information from high-level
network feature maps. The feature map's output dimension
is N N [3 (4 + 1 + M)], where N N is the number of output
grids, each of which has three anchor boxes,a 4-dimensional
prediction value box with the values tx, ty, w, and th, a 1-
dimensional object prediction box with the confidencelevel,
and M dimensions of object category numbers. YOLOv3
predicts the categories during training by using a binary
cross-entropy loss function and numerous logical classifiers
in place of SoftMax to classify each box.
Fig 1.YOLO V3 Architecture](https://image.slidesharecdn.com/irjet-v9i7213-221022093642-411b7583/85/Safety-Helmet-Detection-in-Engineering-and-Management-2-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1125
Fig 2. Architectural Design
3. RESULT AND DISCUSSION
Several prior related studies were used to evaluate the
performance of the proposed systems. The accuracy ranges of
the previous research are between 96.8 percent and 66.3
percent.
Fig 4. Labor Registration Page
Fig 5. Helmet Detection Page
4.CONCLUSION
The approach for identifying helmet use proposed in this
research is based on the upgraded YOLO-v3. It performs the
helmet wearing detection test using data from the
construction site video and web crawler images. Then, to
ensure more accuracy, enhance the YOLO v3 network
utilising target frame dimensiona l clustering and multi-
scaling techniques. The task of helmet wearing detection in
the actual industrial environment may be completed with
accuracy and in real-time thanks to this system'shighspeed.
We used a web camera to complete this research, which will
allow us to increase the accuracy of detection in low light
situations in the future.
5.REFERENCE
[1] Redmon J, Divvala S, Girshick R, et al. You only look
once:unified, real-time object detection//Proceedings
of the IEEE Conference on ComputerVisionandPattern
Recognition, 2015:779-788.
[2] Redmon J, Farhadi A.YOLO9000:better, faster,
stronger/IEEE Conference on Computer Vision and
Pattern Recognition, 2017:6517-6525.
[3] Redmon J, Farhadi A. YOLOv3: an incremental
improvement[J].IEEE Conference on Computer Vision
and Pattern Recognition, 2018:89-95.
[4] YOLOv4: Optimal Speed and Accuracy of Object
Detection. BOCHKOVSKIY A,WANG C Y,LIAO H Y M.
https://arxiv.org/abs/2004.10934.2020
[5] Yan Rongrong. Research on safety helmet detection
algorithm for industrial scene [D]. Xi'an University of
Technology, 2019.
[6] Feng Guochen, Chen Yanyan, Chen Ning, Li Xin, Song
Chengcheng. Research on Automatic Identification
Technology of Safety Helmet Based on Machine
Vision[J]. Machine Design and Manufacturing
Engineering, 2015, 44(10): 39-42.](https://image.slidesharecdn.com/irjet-v9i7213-221022093642-411b7583/85/Safety-Helmet-Detection-in-Engineering-and-Management-3-320.jpg)
Safety Helmet Detection in Engineering and Management
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1123 Safety Helmet Detection in Engineering and Management Jyothika R1, Shweta Salapur2, Mrs. Swathi Sridharan 3 Font Size 12 1,2 Student Dept. of Information Science and Engineering, BNM Institute of Technology, Karnataka, India 3Assistant Professor, Dept. of Information Science and Engineering, BNM Institute of Technology, Karnataka, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract— Due to a lack of knowledge about safety helmets, accidents and injuriesonconstructionsitesarenow increasingly common. Worker supervision by hand is challenging and ineffective. This study is visually checking the construction site to see if anyone is wearing a safety helmet, and then notifying the worker and manager with a sound BUZZER and an SMS if they are. In order to recognise a safety helmet in real time at a building site, we built a deep learning-based technique. Helmet detection was done, and the experimental results indicate that less than 90% of people wore helmets.Comparedtopreviouslyusedmethods, our model has an accuracy rate of 92%. The YOLO-V3 algorithm, which is based on convolutional neural networks, is used in the method that is being discussed. Key Words: Accidents, BUZZER, SMS, YOLO-V3, Deep Learning 1.INTRODUCTION Currently, complex structures and industries are expanding quickly over the globe, requiring a large workforce on construction sites. Unexpected accidents and injuries are more likely to occur on a construction site because of the complicated environment. In order to prevent these occurrences, safety helmets are required in this region. A significant amount of unstructured image data is available on-site thanks to thevideomonitoringequipment.According to the State Administration of Work Safety's accident statistics from 2015 to 2019, of the 80 construction accidents that were reported, 57 occurred as a result of improper use of safety helmets by the workers, making up 69 percent of the total. number of mishaps on construction sites, traditional safety helmet wear checking is quite challenging and requires a lot of manual labor. Additionally, using a visual monitor forces inspectors to spend a lot of time staring at the screen, which can be ineffective. On construction sites, researchers create deep learning-based ways to identify safety helmet use, which can help prevent accidents, injuries from false alarms, and reduced accuracy rates. This study used the real-time object detectionmethod YOLO-V3 (You Only Look Once). utilises deep convolution neural network characteristics.Yolo-V3hastheadvantageof being a lot faster and more accurate than other networks. The main goal of this study is to use live streaming to monitor the building site, identify any workers who are not wearing safety helmets, and notify them via SMSand buzzer. 2. RELATED WORKS To prevent injuries to workers at the construction site, it is crucial to keep an eye on the area and be promptly alerted if someone is not wearing a safety helmet. According to physiologic research, people are only capable of monitoring two signals at once with an accuracy rate of less than 70% [1]. A multidisciplinary field related to computer vision, pattern recognition, signal processing, communication, embedded computing, and image sensor is known as intelligent multi camera video surveillance. The scales and complexities ofcamera networksaregrowingassurveillance technologies advance quickly, and the monitored environments are getting more complex and denser [2]. A video surveillance application that automatically analyses a motorbike rider's helmet use shows promise because helmets are crucial for preventing brain injuries in traffic accidents. In order to segment the objects, a Gaussian-N mixing model is used (GMM). In order to recognize motorcycles in the foreground objects that have been labelled, the suggested system then adapts a faster region- based convolutional neural network (faster R- CNN) [3]. Recently, it was recognized that employing a Convolutional Neural Network (CNN) or Deep Learning, digital image pattern recognition and feature extraction have been successful over the years. The effectiveness of utilizing a convolutional neural network for feature extraction and pattern detection in digital images. Wearing a helmet before entering the workplace is a requirementforthefactorysince the environment of the steel industry workshop is complicated and there may be unanticipated dangers. Helmet testing is a crucial component of the intelligent monitoring system for steel plant staff, since it allowsfor the monitoring of this condition. Accuracy of quicker RCNN algorithms decreases in dim light and complex backgrounds [5]. employing object segmentation and background subtraction in the surveillancefootage. Usingvisual cuesand binary classification, it then establishes whether the bike rider is wearing a helmet or not. Results from the experiment show that the accuracy for detecting bike riders and violators is 98.88 percent and 93.80 percent, respectively. A frame is processed on average in 11.58 milliseconds, which is suitable for real-time use[6]. Thetask of automated motion detection in traffic monitoring is difficult. This study develops a technique to get meaningful data from security cameras for tracking moving objects in digital films. The outcomes demonstrate that the suggested method recognises and tracks moving objects in urban
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1124 movies with success [7]. The majority of motorcycle riders are seen to drive without wearing helmets. In most cases of vehicle accidents, wearing a helmet can lower the risk of head and severe brain injuries to motorcycle riders. It can only be used to identify helmet wea rings; it cannot be used to detect triple riding on a motorcycle [8]. 3. PROPOSED METHODOLOGY The project has been implemented by dividing the entire project into three modules. They are: Data Collection Data Pre-processing Detection Of Helmet A. DATA COLLECTION In this module weare collecting livefacedatasetforface recognition for every person we are collection 200 images to train the model. Data collection is the process of gathering information on specific areas with the a I’m of evaluation. In this study each trained color image is converted as gray image. B. DATA PREPROCESSING In this module we are applying image resizing and normalizing the image using OpenCV. And building the variousLBPHalgorithmforfacerecognition. Pseudo code: Procedure Building Model () Input: Cleaned Data Output: Pre-Trained Model Step 1: Read the dataset using CV2 Step 2: Extract the face features Step 3: Convert into Numerical Array Step 4: Build model Step 5: Train Model using data Step 6: Save the pre-trained model C. DETECTION OF HELMET In this module we are taking input camera and loading the yolov3 model to identify the people and count the persons inthe frame then find outthey are wearinghelmet or not, if they are not wearing helmet system will send a lert SMS and make BUZZER. Pseudo code: Procedure Helmet Detection () Input: Camera frame Output: Alarm if no helmet Step 1: Read frame Step 2: Load pre trained model Step 3: Count person Step 4: For all the person Step 5: Detect the helmet Step 6: If helmet detected continue Step 7: else Step 8: Identify the face and send alert sms Step 9: Return result D.ALGORITHM The classifier YOLO-V3 In contrast to YOLOv2, the pooling layer is cancelled in the YOLOv3 network, which only has a convolutional layer. The size of the output feature map can be altered by varying the convolutional layer's stepsize.The YOLOv3 network employs the DARKNET53 feature extraction network. YOLOv3 employs a tinyfeaturemapand incorporates the concept of a "pyramid feature map" (FPN). The usage of large feature maps allows for the detection of large objects. The rationale for this strategy is that the higher-level network features include richer semantic information while the lower-level network features have more accurate location characteristics. The YOLOv3 prediction layer is able to precisely find the object and correctly classify itbecausetothefeaturemap'scombination of accurate location information from low-level network feature maps and rich semantic information from high-level network feature maps. The feature map's output dimension is N N [3 (4 + 1 + M)], where N N is the number of output grids, each of which has three anchor boxes,a 4-dimensional prediction value box with the values tx, ty, w, and th, a 1- dimensional object prediction box with the confidencelevel, and M dimensions of object category numbers. YOLOv3 predicts the categories during training by using a binary cross-entropy loss function and numerous logical classifiers in place of SoftMax to classify each box. Fig 1.YOLO V3 Architecture
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1125 Fig 2. Architectural Design 3. RESULT AND DISCUSSION Several prior related studies were used to evaluate the performance of the proposed systems. The accuracy ranges of the previous research are between 96.8 percent and 66.3 percent. Fig 4. Labor Registration Page Fig 5. Helmet Detection Page 4.CONCLUSION The approach for identifying helmet use proposed in this research is based on the upgraded YOLO-v3. It performs the helmet wearing detection test using data from the construction site video and web crawler images. Then, to ensure more accuracy, enhance the YOLO v3 network utilising target frame dimensiona l clustering and multi- scaling techniques. The task of helmet wearing detection in the actual industrial environment may be completed with accuracy and in real-time thanks to this system'shighspeed. We used a web camera to complete this research, which will allow us to increase the accuracy of detection in low light situations in the future. 5.REFERENCE [1] Redmon J, Divvala S, Girshick R, et al. You only look once:unified, real-time object detection//Proceedings of the IEEE Conference on ComputerVisionandPattern Recognition, 2015:779-788. [2] Redmon J, Farhadi A.YOLO9000:better, faster, stronger/IEEE Conference on Computer Vision and Pattern Recognition, 2017:6517-6525. [3] Redmon J, Farhadi A. YOLOv3: an incremental improvement[J].IEEE Conference on Computer Vision and Pattern Recognition, 2018:89-95. [4] YOLOv4: Optimal Speed and Accuracy of Object Detection. BOCHKOVSKIY A,WANG C Y,LIAO H Y M. https://arxiv.org/abs/2004.10934.2020 [5] Yan Rongrong. Research on safety helmet detection algorithm for industrial scene [D]. Xi'an University of Technology, 2019. [6] Feng Guochen, Chen Yanyan, Chen Ning, Li Xin, Song Chengcheng. Research on Automatic Identification Technology of Safety Helmet Based on Machine Vision[J]. Machine Design and Manufacturing Engineering, 2015, 44(10): 39-42.