IRJET- A Deep Learning based Approach for Automatic Detection of Bike Riders with No Helmet

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 03 | Mar 2020 www.irjet.net p-ISSN: 2395-0072
S© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 590
A DEEP LEARNING BASED APPROACH FOR AUTOMATIC DETECTION OF
BIKE RIDERS WITH NO HELMET
Pavithra S1*, Priyadharsini M2, Jayalakshmi S3
1,2,3UG Students Department of Information Technology, SRM Valliammai Engineering College, Kancheepuram,
Tamil Nadu, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Detection of traffic rule violators is a
challenging task. It is a critical part of many applications
such as traffic surveillance. Helmet detection plays an
important role in the identification of traffic rule
violators. A method is developed combiningclassification
and cluster for helmet detection. The proposed method
involves Pre-processing, feature extraction, and
classification. It is demonstrated by using surveillance
traffic videos. Finally,themethodwillclassifywhetherthe
person is wearing a helmet or not. Aftertheclassification,
if the person captured is not wearing a helmet it will send
a message with a fine amount to the corresponding
person. As far the robustness and effectiveness are
concerned, this method is betterthanexistingalgorithms.
Key Words: Traffic rule violators, Pre-processing,
Feature Extraction, Classification
1 INTRODUCTION
Two-wheeler is the most convenient and easy mode of
transportation. It is mandatory to wear a helmet in
heavy traffic areas to prevent accidents. By considering
the use of helmet, Governments have made it a
punishable offense to ride a bike without a helmet and
have adopted manual strategies to catch the violators.
Image processing means processing the images based
on the application with the specific parameters. Pre-
processing is the first step to improve the quality of the
images. The feature descriptor algorithm is used to
extract the exact feature andtodifferentiateonefeature
from another. CNN classifier is used to split the images
into two groups, one for training data and another for
test data to use in classification. A ConvolutionalNeural
Network (CNN) is a class of artificial neural networks
used in image processing that is specificallydesignedto
process pixel data.
1.1 OBJECTIVE
The main aim of this project is to detect the bikers with
no helmet, without manual interference and also detect
the
license number plate of the motorcycle. It alerts the
person through phone number with fine amount. This
will prevent road accidents.
1.2 OVERVIEW
The Helmet detection system is recommended for the
identification of a particularpersonwithnohelmet.The
input to the system is captured video which is then
converted into images. Then preprocessing functions
are applied to the image such as background noise,
enhancing contrast and binarization of images.Inorder
to know the characteristics of the image, the Feature
descriptor algorithm is used to extract the exactfeature
and to differentiate one feature from another. CNN
classifier is used to splittheimagesintotwogroups,one
for training data and another for test data to use in
classification. After extracting the Region of Interest
(RoI), the CNN classifier is being trained by a certain
number of pictures wearing a helmet is provided. By
matching RoI and trainedfeatures,itwillbedetermined
whether motorcyclists are wearing a helmet or not.
Convolutional Neural Network is used to solve the
classification problem efficiently.
1.3 APPLICATIONS
The main application of helmet detection is to prevent
accidents in traffic areas. Even though the government
takes various measures, it is not properly followed by
themotorcyclists,soseveralsmarttechniquesshouldbe
employed. Construction industry and powersubstation
suffer a lot of difficulties because of carelessness in
wearing safety helmets. Hence, there is a need for a
surveillance system that is capable of detectinghelmets
and preventing the deaths. A more sophisticated
computer vision model that encompasses image
processing, machine learning, Convolutional neural
networks (CNN), classifiers such as support vector
machine (SVM), ViBe background modeling algorithm,
Histogram of Oriented Gradients (HOG) features and
other techniques will solve the problem.

2. LITRATURE SURVEY
The systems have encountered many reference papers,
this work enabled to understand a deep learning-based
approach for helmet detection of bike riders with no
helmet in a better way.
[1] Felix Wilhelm Siebert et al, “Detecting
motorcycle helmet use with deep learning”-
Elsevier, 2019.
The helmet detection is performed by Object detection
algorithm YOLO9000. It involves annotation for
sampling the video clips and RetinaNet.
DRAWBACKS
 It needs more data for helmet detection
accuracy in motorcyclists with more than two
drivers.
 Diverse video data to be collected according to
the camera angle.
[2] Kunal Dahiyaetal,“Automaticdetectionofbike-
riders without helmet using surveillance videos in
real-time”-InternationalJointConferenceonNeural
Networks (IJCNN), 2016.
The detection of a helmet from the bike riders that
could be performed in this Automatic detection system
by using visual featuresandbinaryclassifier. Histogram
of oriented gradients (HOG), scale-invariant feature
transform (SIFT), and local binary patterns (LBP) are
the three main performance comparison method.
DRAWBACKS
 Video surveillance-based methods are passive
and require significant human assistance.
 It is not an efficient solution due to its
requirement of dedicated hardware.
[3] Archana D et al, “Mission on! Innovationsinbike
systems to provide a safe ride based on IOT”- 2nd
International Conference on Computing and
Communications Technologies (ICCCT), 2017.
In case of making machines more sophisticated in their
way of learning and making decisions we develop the
intelligenceapplication.Toincreasesafetywhiledriving
we implemented this method. The bike machine has
started only when the person wears a helmet. By
making the handlebar vibration to intimate the over
speed performed by the user.
DRAWBACKS
 Cost is the major hindrance to the widespread
use of safety systems.
[4] Rongbao Chen et al, “An Improved License Plate
Location Method Based on Edge detection”-
International Conference on Applied Physics and
Industrial Engineering, 2012.
In license plate detection, the location of the license
place is more important. Using the Prewitt arithmetic
operator that identifies the license plate even under
different backgrounds and lighting conditions by
preprocessed plate image.
DRAWBACKS
 More similar database is needed to compare all
edge-based methods.
[5] Sarbjit Kaur et al, “An Automatic Number Plate
Recognition System under Image Processing”-
International Journal of Intelligent Systems
Technologies and Applications, 2016.
Using the computer vision and image processing
technology, the number plate that has been detected
automatically and also extracts the number plate from
the whole vehicle image. The vehicle that can be
preprocessed first by iterative bilateral filtering and
adaptive histogram equalization.
DRAWBACKS
 Bad weather and hindrances can make
automatic license plate recognitionsystemsnot
completely effective.
3. MODULE DESCRIPTION
(i) INPUT VIDEO DETECTION
The input video has been captured by using either
ipcam or webcam. It is thenconvertedintoimages,from
this face of the biker is identified by using the haar
cascade classifier algorithm to detect whether the
person is wearing a helmet or not.
(ii) IMAGE CLASSIFICATION
Image classification involves converting the captured
images into a binary image, grayscale image and colour
image for further classification. After the conversion of
images, it then compared with trained images in the
database for evaluation.

Fig -1: Images stored in greyscale
(iii) CNN CLASSIFIER
Convolutional Neural Networkis a partof deep learning
neural network. It contains special architecture with a
multi-layeredneuralnetworktodetectcomplexfeatures
in data. CNN contains a feature descriptor called kernel
or filter, which converts the images into matrix
representation to identify complex features.
Fig -2: CNN Operation
(iv) RESULT INTERPRETATION
In the final step, after the classification of images and
comparison of images with trained images, the system
detects whether the person is wearing a helmet or not
and shows the result. If the person is not wearing a
helmet, it captures the license plate of the bike and
generates SMS with a fine amount.
4. SYSTEM DESIGN
In System design, the language used is Python. The
backend process is OpenCV and dataset.
4.1 ARCHITECTURAL DESIGN
The proposed system involves feature descriptors and
neural networksforhelmetdetection.Thefirststepisto
capture the video input for face detection and then it is
convertedintoimagesforclassification.Thesecondstep
in the detection process is Pre-processing, which
involves enhancing the important features like image
contrast, pixel brightness, geometric transformation
and removal of distortions for further processing. After
Pre-processing the next step is feature extraction in
which the various features of the image are extracted
using the feature descriptor algorithm. In the final step,
the trained images in the database are compared with
captured images forfurtherclassification.Afterthefinal
classification, it shows whether the person is wearing a
helmet or not.
Fig -3: Block Diagram
4.2 PROPOSED METHOD
Image procurement: The input image is captured
through ipcam or webcam. It is the major part of any
vision system.
Preliminary processing techniques: This step is
mainly focused on removal of background noise,
enhancing contrastness and binarization of images.
Helmet detection: After extracting the Region of
Interest(RoI), classifier which is being trained by a
certain amount of picture wearing helmet is provided.
By matching RoI and trained features, it will be
determined that whether motorcyclists is wearing
helmet or not.
License plate detection: If the person didn’t wear a
helmet, it captures the license plate of the particular
vehicle.
SMS generation: After capturing the license plate, it
generates the SMS to the person with a fine amount.
CNN Technique
Convolution is the integration of two functions that
shows how one function modifies the other. The
formula for convolutional function is,

The three important items in this process are the input
image, the feature detector and the feature map. The
input image is the image detected from the camera. The
feature detector is a 3x3 matrix and is also referred to
as a kernel or a filter. The input image is represented in
the matrix and it is multiplied element-wise with the
feature detector to produce a feature map. The feature
map is also known as the convolved feature or an
activation map. The main aim of the convolution is to
reduce the size of theimageandmakeprocessingfaster.
Fig -4: CNN Features
5. OUTPUT
Outputs obtained are as following:
1. Person wearing the helmet is detected.
Fig -5: Helmet Detection Process
2. After detecting the helmet, the following message is
displayed.
Fig -6: Helmet Detected Message
3. If the person didn’t wear the helmet, the license plate
is captured.
Fig -7: License Plate Captured
4. After capturing the license plate, SMS is generated.

Fig -8: SMS alert
5. CONCLUSION
In the proposed system, the existing algorithms such as
support vector machine (SVM), ViBe background
modeling algorithm, Histogram of Oriented Gradients
(HOG) is replaced by CNN techniques. A framework is
represented for detection of traffic rule violators who
ride bike without using helmet. The proposed system
provides efficiency and robustness. Convolutional
Neural Network is a part of deep learning neural
network. It contains special architecture with a multi-
layered neural network to detect complex features in
data. CNN contains a feature descriptor called kernel or
filter, which converts the images into matrix
representation to identify complex features.
5.1 FUTURE WORK
In future, for a more reliable and less complex system
the system can be improved by substituting advanced
techniques. This application can be implemented in
various real-time scenarios like Video Surveillance,
Outdoor Object recognition systems, Security Process
on Toll Gate etc.
REFERENCES
[1] Felix Wilhelm Siebert et al, “Detecting motorcycle
helmet use with deep learning”- Elsevier, 2019.
[2] Kunal Dahiya et al, “Automatic detection of bike-
riders without helmet using surveillance videos in real-
time”- International Joint Conference on Neural
Networks (IJCNN), 2016.
[3] Archana D et al, “Mission on! Innovations in bike
systems to provide a safe ride based on IOT”- 2nd
International Conference on Computing and
Communications Technologies (ICCCT), 2017.
[4] Rongbao Chen et al, “An Improved License Plate
Location Method Based on Edge detection”-
International Conference on Applied Physics and
Industrial Engineering, 2012.
[5] Sarbjit Kaur et al, “An Automatic Number Plate
Recognition System under Image Processing”-
International Journal of Intelligent Systems
Technologies and Applications, 2016.

IRJET- A Deep Learning based Approach for Automatic Detection of Bike Riders with No Helmet

More Related Content

What's hot (19)

Similar to IRJET- A Deep Learning based Approach for Automatic Detection of Bike Riders with No Helmet (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- A Deep Learning based Approach for Automatic Detection of Bike Riders with No Helmet