HELMET DETECTION AND LICENSE PLATE RECOGNITION

A.Vandana Peter, Ch. Sai Lakshmi, K. V. N. L. Prathyusha, P. Anil Kumar, Ch. Hemanth Manikanta
CSE, Godavari Institute of Engineering and Technology, Rajahmundry, Andhra Pradesh.

Abstract: Motorcycles are widely used in I. INTRODUCTION

developing countries as the essential mode of
transport. In last recent years, there has been a
All over the world many lives are lost every year
rapid increase in road accidents owing to the fact
due to road accidents, and millions of people are
that majority of the motor bicyclist fail to wear
getting injured due to road accidents and they are
helmet that makes it an ever-present danger. In
becoming disabled due to their injuries. It is hard
this decade, most of the accidents are caused
to imagine that burden is unevenly borne by
because of the head injury. Due to this, wearing
bicyclists, cyclists and pedestrians. The most
helmet is formed necessary by means of traffic
worrying fact is that India ranks number one as far
rules. But most of the motorbike riders never
as road crash deaths are considered. The main
obey the rule. Many cities make use of a
reasons are due to not wearing of helmet, seatbelt,
surveillance network to monitor the bicyclists
violating traffic rules behind this accidents
violating the helmet laws. But such a system will
according to analysis done by experts.
need human intervention. The surveys say that
Policymakers first acknowledge the problems that
human interventions prove ineffective, due to the
persists in India before halving road accident
increase in the time of monitoring and also due to
deaths. When a motorcyclist meets with a road
the errors made by human during monitoring.
accident, due of sudden brakes, the rider is thrown
The main idea of this application is to detect the
away from the vehicle. If head strikes any object,
license plate number of those who violate the law
motion of the head becomes zero, but with its own
of wearing helmet using Deep Learning
mass the brain continues to be in motion until the
Algorithms.To detect the bicyclists who are
object hits inner part of the skull. Many times this
against the helmet laws, a system which uses
type of head injuries may be fatal in nature. In
convolutional neural network and image
such times helmet acts as life savior. Helmet
processing is implemented. This system deals
reduces the chances of skull getting decelerated,
with the detection of motorbike, Helmet or Non-
hence sets the motion of head to almost zero.
Helmet classification and License plate
Cushion inside the helmet absorbs the impact of
recognition of Motorbike. YOLOv3 feature is
collision and as time passes head comes to a halt
used to detect the motorbikes. When the bike is
and is dead. It also spreads the impact to a larger
detected using convolutional neural
area, thus safeguarding head from the severe
network(CNN), it determines whether the
injuries.
bicyclist wears a helmet or not. If the rider is
Deep learning is one of the machine
identified with no helmet, then the license plate
learning techniques that teaches the computer to do
number of his motorcycle is captured and
works as naturally as humans. The key technology
extracted using Tesseract OCR.
behind driverless cars that enables them to
recognize the stop sign, or to distinguish a
pedestrian is Deep Learning. It works as the key to
Keywords: Deep Learning, Yolov3,
control voices in consumer devices like TVs,
Convolutional Neural Network, Tesseract OCR.
phones, tablets. This technology, Deep learning is [6], C. Vishnu, used Convolutional Neural
getting lots of attention recently for a good reason. Network(CNN )for his classification. In this
It helps in achieving results that were absolute and proposed model he created the dataset from the
accurate. surveillance network of videos. This system helps in
Helmet Detection and License Plate Recognition.
II. LITERATURE SURVEY [7] B. V. Kakani , proposed his recent work to
detect the number plate system. He used Optical
Character Recognition(OCR) using neural
[1] , Chiverton, proposed a system for the
networks.
automatic detection of bicyclists who doesn’t wear
Helmet. In his research work he trained the images
by taking the head portions of bike riders using a III. METHODOLOGY
SVM classifier. The main features taken into
consideration are the shape and reflective property This system tries to detect whether bicyclists
of helmets which looks brighter. This method leads wearing helmet or not, if the rider doesn’t wear the
to lots of misclassifications as the objects with helmet then the system extracts the license number
similar shape and bright surfaces are considered to plate of that vehicle. To extract the license plate
be helmets. Also, this method did not identify the number plate we have YOLO CNN model with
motorcycles. some train and test images and if you want to add
[2] , Silva et al, proposed a method to resolve the some other images then send those images to us, so
misclassification by training the system to classify we can include those images in YOLO model with
the motorcycles from frames by extracting the local annotation to extract number plate of those new
Binary Pattern descriptor using SVM classifier. images. Object Detection is the main principle
Then tried to classify helmet using circular Hough involved in this methodology using Deep Learning
transform, local binary patterns and histogram of at three levels. More importantly it acts as a barrier
gradient. between the head and the object to which the rider
[3] , Doungmala et al, proposed to system to detect comes into contact. Injuries and accidents can be
half and full helmet using decision tree classifier minimized if a good quality full helmet is used and
with AdaBoost. In this method, it first uses the helmet laws are followed.
features of Haar and circular Hough transform to . Traffic rules are there to bring a way of discipline,
detect face and classify it as with or without helmet in order that the danger of deaths and injuries are
and full or half helmet. But, it has its own often minimized significantly. However strict
limitations as it acquires the video from the traffic. adherence to those laws is absent actually . Manual
This proposed model also failed to detect the surveillance of traffic using CCTV, an existing
motorcycle. methodology can be replaced with the proposed
[4], K. Dahiya, proposed a model that detects model.
helmet from the survelliance videos. This system In this system, a Non-Helmet bicyclist who violates
uses a SVM classifier for the classification of the traffic rules of not wearing helmet is detected
motorcyclist and non-motorcyclist. It also classifies and the license plate number of his vehicle is
helmet and non-helmet using another SVM detected automatically. Object Detection is the main
classifier. Features like HOG, SIFT and LBP are principle of this system which uses Deep Learning
used to implement these both classifiers. When at three levels. This system detects the person,
compared with other two features, HOG descriptor motorcycle as objects at first level using YOLOv2
gives best performance. model, the helmet at second level of process using
YOLOv3 model, License plate of the vehicle at the
last level of the system using YOLOv2 model. The Character Recognition(OCR). Injuries and accidents
license plate number extraction is done using can be minimized if a good quality full helmet is
Optical Character Recognition(OCR). Since, this used and helmet laws are followed.
system takes both images and videos as its input, . Traffic rules are there to bring a way of discipline,
the speed of execution plays a crucial role. All the in order that the danger of deaths and injuries are
above methodologies are used to build a holistic often minimized significantly.
system that detects helmet and extracts the license Advantages :-
plate number. • Injuries and accidents can be minimized if a good
quality full helmet is used and helmet laws are
followed.
• For accuracy improvement of helmet detection
PCA technique is employed.
• For accurate license plate detection and character
extractions several methods are used such as
Inception-v3, OCR.

V. RESULTS

In this section, the user need to upload the image or

video to detect the License numbers of bikes when
the riders violate the law of wearing Helmet.

Upload the image

Architectural Model

IV. PROPOSED SYSTEM

In this system, a Non-Helmet bicyclist who violates

the traffic rules of not wearing helmet is detected
and the license plate number of that vehicle is
detected automatically and license plate number is
extracted. Object Detection is the main principle of
this system which uses Deep Learning at three
levels. This system detects the person, motorcycle
as objects at first level using YOLOv2 model, the
helmet at second level of process using YOLOv3
model, License plate of the vehicle at the last level
of the system using YOLOv2 model. The license
plate number extraction is done using Optical
 Upload Video
Detect Person

Detects Helmet with accuracy

Detects Helmet with accuracy

Detects the Non-Helmet Riders

Extracts the License Plate Number if Helmet is
not detected
 VI. CONCLUSION 5. Kunal Dahiya, Dinesh Singh, C. Krishna Mohan,
“Automatic Detection of Bike-riders without Helmet
A Non-Helmet Bicyclist Detection and License number using Surveillance Videos in Real-time”, International
recognition system is developed where a video file and joint conference on neural network(IJCNN) , 2016.
image file is given as input. If the motorcycle rider in
the video and image violates the helmet law of not
wearing helmet while riding the bicycle, the license
plate number of that motorcycle is extracted from
image and displayed. Object detection with YOLO
architecture is used to detect the motorcycle, person,
helmet and license plate number. OCR feature is used
for the extraction of license plate number if rider fails
to wear helmet.The frames and the characters
recognized from this feature can be used for other
purposes. The objectives of this project work is
achieved satisfactorily.
The accuracy obtained for motorcycle or non-
motorcycle classification is 93%, helmet or non-helmet
classification is 85% and the license plate number
recognition is 51% resulting in an average accuracy of
around 76%.The accuracy of this system can be
increased by improving the training data set of videos,
images and by increasing the image quality.

VII. REFERENCES

1. Thepnimit Marayatr, Pinit Kumhom, “Motorcyclist’s

Helmet Wearing Detection Using Image Processing”,
Advanced Materials Research Vol 931- 932,pp. 588-
592,May-2014.
2. J. Chiverton, "Helmet presence classification
with motorcycle detection and tracking", IET
Intelligent Transport Systems, vol. 6, no. 3, pp.
259-269, September 2012.
3. Abu H. M. Rubaiyat, Tanjin T. Toma, Masoumeh
Kalantari-Khandani, “Automatic Detection of Helmet
Uses for Construction Safety”,WIC,ACM International
Conference on Web Intelligence Workshops (WIW),
2016.
4. XINHUA JIANG “A Study of Low-resolution Safety
Helmet Image Recognition Combining Statistical
Features with Artificial Neural Network”. ISSN: 1473-
804x