Topic : Helmet Detection System using Machine Learning.

Group Members
Under the guidance of Dr. Mahesh Chandra Prof.,
Vishal (R18EC350)
School of ECE
Varun Bander M (R18EC336)
REVA University
Thippeswamy MM (R18EC330) Arun
Kumar S (R18EC365)
Group No: 64
Group Name: Project Z
Contents.
• Abstract.
• Introduction.
• Literature survey.
• Objectives.
• Methodology.
• Implementation.
• Applications.
• References.
Abstract.
The project deals with a method for detecting motorbike riders with & without a helmet with
the help of machine learning. Two-wheeler accidents are increasing gradually in all
countries. A helmet or a protective cap is the predominant safety equipment of two-
wheelers, motorcycle riders, and pillion riders, however many drivers neglect wearing
helmets. The main outcome of wearing a helmet is to protect the head of a person traveling
on two-wheelers just in case of a major or minor accident or fall from a running bike. We
worked on a new approach that priorly collected a dataset of images and videos of road
traffic where we collected various kinds of photos like with driver wearing a helmet, a
driver without wearing a helmet and also the primary rider is with a helmet and second
person without a helmet so and differentiated the 2 wheelers from other heavy and light
vehicles like cars and bus on road. This detection of wearing a helmet is done through
machine-learning and utilizing a free open-source library cross-platform using which we can
develop real-time computer vision applications software called OpenCV.

3
Introduction.
• In every country around the world, two-wheelers, bikes are most utilized for transport.
• A middle-class family or small family with few members utilizes a bike as their primary
transportation. This is due to the lesser price and very few maintenance charges. But the problem
raised by the two-wheelers is less security, unproductive, and high risk is entangled with bikes.
• It is suggested to always wear a helmet while riding two-wheelers.To minimize the involved risk of
life, it is highly recommended and encouraged for bicycle riders to wear a protective cap or helmet.
• Nowadays there are pre-existing strategies that use sophisticated sensors fixed to the motorbike to
evaluate whether the bike rider is using a helmet and the bike engine starts only if they wear a
helmet.
• But, it is not possible to set the mind of each person to mount sensors on the bikes because of the
cost involved in the process of fitting these sensors to bikes.
• As in this technologically up to date world, in almost all towns, cities a huge video observation is
implemented for surveillance on a wide diversity of street dangers and traffic monitoring.
• In this current situation utilizing such a pre-existing method is a cost- efficient system, real-time
problem solving is possible like traffic violators and proof for any mishaps, accidents, or can also be
employed to the identification of bike riders without a helmet or a security cap .
4
 LITERATURE SURVEY.
Serial Number Title Authors Year of publication Limitations About

1. A Hybrid Approach for Helmet M.Swapna, 2017. Capturing bikes with high A list of all bike riders
Detection for Riders Safety. Tahniyath, speed are difficult to find driving without wearing a
Wajeeh, here. helmet along with a
Shaziya Jabeen. snapshot for proof.

2. Helmet Detection on Motorcyclists Romuere 2018. Limitations of low quality This study addresses the
using image descriptors and R.V.Silva, images can not be able to detection of motorcyclists
classifiers. Kelson, generate promising results without helmets on public
M.S.Veras, roads.Techniques used are
R.T.Aires. CL,WT,MPL
algorithm.Results obtained
are vehicle segmentation
and classification of helmet
use.

3. Automatic detection of bike riders Kunal Dahiya, 2016. It tries to locate a helmet HOG,SIFT,LBP,
without helmet using surveillance Dinesh Singh, in full frame which is RBF are the techniques
videos in real time. C.Krishna computationally used in this paper.
expensive,so it is confused
for other similar objects as
helmet .
Serial Title Authors Year of Limitations About
number Publications
4. Helmet Ramesh Babu, 2018. The Techniques This does not
Detection on Amandeep used are work there are
two wheeler Krishnangini,M Machine n multiple
riders using ahima Singh. Learning,Super vehicles in the
machine vised scene and
learning. Learning,Featu accuracy is
re Extraction. very low.

5. Helmet Presence J.Chiverton. 2012. Results of the Techniques

classification tests were using the
with motorcycle unsatisfactory,t support vector
detection and hey produced Machine are
tracking. too many false used as a
positives and classifier.
false negatives.
Objectives.
• To ensure a safe ride to the rider.
• In case a person riding a vehicle meets with an accident, there are high chances
of them damaging their heads. Wearing a helmet provides protection in such
situations.
• Wearing a helmet is a mandatory traffic rule and breaking it may lead to paying
fines. To avoid this situation, it’s better to detect the wearing of a helmet and let
the user know in advance.
Methodology.
The proposed work involves in implementation of two important steps:
i) step 1 is an implementation of a neural network which is implemented for determination
of mono and couple of riders on a bike by the sophisticated and up to date YOLOv3(you
look only once) model .
ii) We adopted one of the deep neural networks in the offered framework which is
responsible for the spotting of motorcycle riders, the second person sitting on the bike
whether they are using a helmet. In order implement this framework, traffic video is the
must to the YOLO3 model to process through the two steps, and then single video frames
are obtained by taking screenshots from processed videos and crop them near the head of
the motorcycle rider is visible and these images are throughput of the next stage that is a
convolution neural network (CNN) to which identifies and centralizes bike riders and
pillion riders wearing or not wearing a helmet or any protective cap.
Block Diagram.
Flow Chart/ Algorithm.
Hardware Description.
This project makes use of no hardware components.

Software Description.
• Convolution Neural Network (CNN)
• OpenCV
• Python
IMPLEMENTATION.
• Assuming there are 1000 photos, So, in order to detect, these are all similar to web
images. As a result, it is required to feed these photographs into our machine learning
model as a series. For example, when the image is loaded, we cannot determine which is
a helmet and which isn't. It must be trained on the computer that the person wants to
wear a helmet. As a result it is required to have a XML file which recognises the helmet.
Each XML file will have a structured name, which will indicate where and how many
helmets there are. As a result, it's truncated, just like the helmet. In this range, it has a
bounding box. To detect a person wearing a helmet, and without a helmet, we require a
min Max bounding box model. As a result, there are two separate files here, one for
photographs of each different helmet and we consider taking into account all different
sorts where someone is not wearing a helmet. One should evaluate all of these distinct
types of photos before developing them. As a result, there is an XML file on each of
these photographs. The helmet and its location of the coordinates on the 2D plane are
important because it is important to know whether the person is wearing a helmet or not,
and if so, where he is wearing it. As a result, Tensor flow will be used, in Tensor flow , a
single sequential neural network layer will be used.
Applications.
• Detection of helmet or without helmet of two wheelers.
• To ensure the safety of the rider.
• To avoid deaths during road accidents.
References.
1. Prasad, M.V.D. &Inthiyaz, Syed &Maddala, Teja & Kanapy, Hari & Manohar, Minna & Kumari, Rupa
&Shaik, Hasane. (2019). Human activity recognition using Deep Learning, International Journal of
Emerging Trends in Engineering Research. 7. 536-541.10.30534/ijeter/2019/227112019.
2. Kishore, P.V.V. n Kumar, K & Puri, Kiran & Sastry, A & Maddala, Teja &Anil Kumar, D. & Prasad,
M.V.D..(2018). Indian Classical Dance Action Identification and Classification with Convolutional
Neural Networks. Advances in Multimedia. 2018.10.1155/2018/5141402.
3. G. Ross, D. Jeff, D. Trevor, and M. Jitendra, ‗Rich feature hierarchies for accurate object detection and
semantic segmentation,‘ in Proceedings of IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), pp. 580–587, 2014.
4. R. Silva, K. Aires, T. Santos, K. Abdala, R. Veras, and A. Soares, ‗Automatic detection of motorcyclists
without helmets,‘ in Proceedings of Latin American Computing Conference.
5. J. Chiverton, ―Helmet presence classification with motorcycle detection and tracking,‖ IET Intelligent
Transport Systems (ITS), vol. 6, no. 3, pp. 259–269,2012.