Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

Helmet and Number Plate Detection

D. Tharun Reddy1; N. Durga Prasad2; M. Sheshu Kumar 3 and Dr. A. N. Satyanarayana4 (Professor)
1 to 3 Students of B. Tech, ECE Department, Sreenidhi Institute of Science and Technology, Hyderabad
Electronics and Communication, SNIST, Hyderabad

Abstract:- An important use of computer technology in The key objective of this work is to prevent
recent years has been the automatic helmet recognition significant injuries caused by motorcycle accidents by
of motorcy clists in real-time surveillance film. Deep applying computer vision-based automatic motorcycle
learning methods are becoming more and more popular helmet recognition. The proposed solution will identify
as a result, especially for object detection and riders who are not wearing helmets in real-time and alert
classification. Nevertheless, a number of issues, including authorities to take appropriate action. This technology can
limited resolution, inadequate lighting, adverse weather, play a vital role in promoting compliance with helmet laws
and occlusion, restrict the accuracy of current models and improving overall road safety.
in identifying motorcycle helmets. A unique method that
makes use of the Faster R-CNN model has been put In summary, the increasing number of motorcycle
out to address these issues. Using the input image as the accidents and head injuries among riders highlight the
starting point, this method first trains the Region importance of implementing measures to promote helmet
Proposal Network (RPN), and then it uses the RPN usage. The pro- posed solution using computer vision-based
weights to train the Faster RCNN model. The goal of this automated helmet detection technology can effectively
method is to increase helmet detection accuracy in live address this issue, par- ticularly in emerging nations with
surveillance footage. This method’s experimental results high motorcycle usage [5]. The primary goal is to prevent
have demonstrated encouraging results, with a 95% significant injuries caused by traffic accidents by promoting
accuracy rate in identifying motorcycle helmets in live compliance with helmet laws through automatic helmet
surveillance footage. This illustrates the promise of deep detection systems.
learning approaches in the field of automatic helmet
detection for motorcyclists in real-time surveillance film,  Problem Definition
as well as the efficacy of the suggested strategy in The main objective of this research is to build an
overcoming the issues encountered by current models. automatic helmet violation detection system with the help of
deep learn- ing approaches of computer vision. The system
Keywords:- Helmet, Faster-RCNN, CNN, Deep Learning, is designed to analyze surveillance videos captured by
Region Proposal Network, Surveillance Videos. cameras installed on roads or in public places and identify
whether motorcyclists are wearing helmets properly or
I. INTRODUCTION violating helmet laws.

The general cause of the rise in motorcycle II. RELATED WORKS

accidents in recent years has been the riders’ haste,
negligence, and recklessness. Since head injuries account Different approaches have been put forth by numerous
for the majority of fatalities in these incidents, preventative researchers to address the issue of autonomous helmet iden-
measures are imperative [1]. A tested safety precaution that tification in real-time traffic situations. The section below
dramatically lowers the chance of brain damage in the event discusses these techniques.
of an accident is wearing a helmet. However, the majority of
brain injuries among motorcycle riders are caused by their In the beginning, researchers employed computer
failure to abide by traffic laws mandating the usage of vision techniques like HOG, SURF, and SIFT for machine
helmets. learning to automatically detect whether motorcycle riders
were wearing helmets or not.
In countries like India, where motorcycle usage is high,
the problem of non-compliance with helmet laws is Many analyses of the traffic on roads, with the
particularly acute. To address this problem, a cost-effective identification, categorization, and tally of cars as well as the
solution using computer vision-based automated systems for detection of helmets, have been carried out in recent years.
helmet detection can be employed [2]. This solution can be The first step in any research on vehicular traffic
easily integrated with existing video surveillance systems development should be to identify and classify the vehicles
and has the potential to significantly reduce the workload that are driven on public roadways. As such, some related
of traffic police. studies are analysed here.

IJISRT24MAY156 www.ijisrt.com 1
Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

A framework for the detection of road safety violators according to the kind of vehicle, were used to calculate these
who ride bicycles without wearing helmets was proposed by models. A variety of versions were built, including ones for
Kunal Dahiya [1]. The suggested framework helps the traffic motorbikes, bicycles, small and large cars, minibuses,
police find these offenders in unusual settings, such the trucks, buses, and pedestrians. For every vehicle that is
blazing sun, etc. The experimental results shows the caught, a model is built to be assessed to the models made
accuracy of 93.80% for violator detection. With an average for every kind of vehicle. The vehicle class is determined by
processing time of 11.58 milliseconds, this data is identifying the model that most closely reflects the type of
appropriate for real-time applications. Additionally, the the seized vehicle. A constraint of this research is that only
suggested structure automatically adjusts, with minimal one design was employed for both motorcycles and bicycles.
tweaking, to new cases as needed. It is possible to expand Furthermore, it has been shown that these kinds of objects
this framework to identify and report violators’ license cannot be well described by categorizing the vehicles based
plates. only on geometrical data. Another problem with this
strategy is that some parameters, such as focal distance,
Chiu C-C [2] suggested a computational vision angle, and camera height, should always have the same
system that aims to monitor and identify motorcycle riders values. A vehicle would not match the models if these
that are partially hidden by another vehicle. The features were changed, hence new ones ought to be created.
approach assumes that the area around the helmet is round According to this research, the system was not able to be
in shape. The limitations 3 of the picture are computed over implemented on certain public roadways due to the
the area around the motorcycle, or the helmet’s potential absence of a location for the camera.
zone, in order to identify the helmet. Then, recursively count
the number of edge points which form a circle. In case this An Approach for MVD Helmet Detection by Rohith C.
number exceeds or equals a current value, the region will be A suggested using Deep Learning. [6]. The chosen condition
associated with a helmet throughout the system calibration for categorizing the recorded frames is a person operating a
process. The system concludes that there is probably a two- wheeler while donning a helmet or not. First, they
motorcycle nearby when it finds a helmet. During the categorize photos that show people riding two-wheelers or
calibration phase of the system, the operator needs to input not. Next, examine the photographs that have been
certain measures like the height, angle of the camera, and classed to get the data set according to the condition. Any
helmet radius. In the event that any conditions change, like captured frames that don’t meet the aforementioned
the height of the camera or the road where the device is requirements for photographs will be ignored or deleted.
being used. The Caffe model is the one that is utilized for extraction
and detection. The accuracy score of the suggested model is
Leelasantitham and Wongseree [3]devised a method 86%. The Inception V3 model serves as the categorization
that uses traffic engineering techniques to detect moving model. The suggested model had a 74% validation accuracy
autos. Five pairs of cars were present. The first category score.
consisted of tricycles, motorbikes, and bicycles. Vans and
sedans were among the cars in the second category. Small Prity Kumari [7] proposed a system which is is
cars and minibuses fall under the third category. The fourth extremely effective for the protection purpose of the user.
group consisted of medium and large buses. The fifth User needs to wear helmet to ride a motorcycle and
category consisted of trailers and large trucks. The features therefore traffic rules are monitored by the rider. This
that were used for classification were the length and width method is below pocket management that’s riding the two
of images. The database consists of just 76 images, wheeler vehicle having safety in hand and in affordable.
casting doubt on the results’ dependability. Remember that This approach uses basic functionalities. It offers the
the system depends heavily on the road where the photos cyclist increased security. Since all of the tools and libraries
are taken. Consequently, if the same system is used on a used in the project are freely available, they are all very
different route, all mea- sures—like the image’s length and adaptable and reasonably priced.
width—should be adjusted. A technique utilizing image
processing to offer data on the number of traffic Addressing the problem of inefficient traffic
offenders in a given area was proposed by Tahniyath organization was the main objective of the project’s
Wajeeh [4]. It snaps a picture as proof and builds an engineering. From now on, the end will state that if
exhaustive list of all cyclists that don’t wear helmets when coordinated by any traffic management departments, it would
driving. Because OpenCV, Tensor Flow, and other open- make their work more efficient and easy. It is demonstrated
source, free technologies are used, the software is by Romuere Silva [8] that the LBP de- scriptor
significantly cheaper. This technique was tested to provide outperformed the HOG and Haar Wavelet descriptors in
complete proof and accurate outcomes under optimal terms of robustness for the task. By pattern joint distri-
lighting circumstances. bution, the LBP descriptor characterizes the regional
texture pattern. Motorcycle texture patterns bolster the
The more people that are aware of the system, the efficiency of the LBP feature-based classifier. The SVM
greater its impact. Messelodi S. presented a method for classifier produced accurate findings. The performance of
vehicle classification and segmentation [5]. To categorize the SVM during the training phase is its primary benefit.
the artifacts, eight 3-D models were developed in order to SVM often trains more quickly than Radial Basis
classify them. The sizes of the vehicles, which differ Function Networks (RBFN) and Multi-layer Perceptrons

IJISRT24MAY156 www.ijisrt.com 2
Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

(MLP). The Random Forest algorithm produced the accurate Adaptive background subtraction is used in C. Vishnu’s
result in the helmet detection step. The collective [14] proposed framework for the automatic detection of
approach of CHT, HOG, and LBP produces an excellent, motorcy- clists who do not wear helmets. This technique is
satisfactory outcome with regard to the hel-met detection. resistant to a number of problems, such as poor lighting and
This makes sense since the feature vector is constructed video quality. A more dependable system is produced by
by combining edge (HOG), texture (LBP), and geometry using machine learn- ing for autonomous learning of
(CHT) insights. discriminating models for task classification, which raises the
rate of identification and lowers the amount of erroneous
A strategy for keeping an eye on motorcyclists who alerts. Using two genuine video records, the trials
choose not to wear helmets is suggested by the GayathriV effectively identified 92.87% of offenders with a small
[9] re- search. A PC visual system including elements for erroneous alert rate of 0.50%, demonstrating the
moving item section, moving item configuration, and helmet effectiveness of the suggested technique.
recogni- tion would be useful to the traffic specialists. The
proposed framework will also assist the traffic police in Jimit Mistry [15] offers an automated helmet detection
finding these of- fenders in challenging circumstances, such technique based on CNN. They use two stage YOLOv2
intense heat, etc. In order to overcome obstacles, propelled models to increase the helmet detection precision. First,
following computations are usually necessary.Night vision human detection is guaranteed by the guaranteed by the
cameras are commonly employed to use the location YOLOv2 model, which used the COCO data collection for
framework when there is no light. Further examples of training. As a result, fewer helmets will go unnoticed. The
both positive and bad behavior will be stored in response to edited photos of the observed person provide input to the
requests to enhance the conjectural capability of the next YOLOv2 model, which was trained using our data set
framework in the future. Work with the front- end video of photos with helmets. The proposed methodology has been
catch modules as well. tested using a range of helmeted image settings.
Additionally, they assessed the numerical measures on the
According to Narong Boonsirisumpun’s research [10], test photos and contrasted them with other cutting-edge
an experiment was carried out to detect and sort techniques. Our approach has a high accuracy in helmet
motorcyclists into groups according to whether or not they detection, and its dependability and resilience are
wear helmets. Four convolutional neural networks (CNNs) demonstrated by experimental results and quantitative
are employed in these tests: VGG16, VGG19, Google Net or measurements on a range of circumstances.
Inception V3, and Mobile Nets. To complete the image
recognition phase, these models are combined using the III. PROPOSED WORK
SSD approach.
The first step in helmet recognition is detecting a
How to spot motorcycle riders who aren’t wearing moving vehicle. It is the first stage before performing more
helmets is covered by Romuere Silva [11]. They propose advanced activities like vehicle tracking or categorization.
dividing a CV system into three components: detection, Rather than processing the full video at once, the example
classification, and segmentation of moving objects. The begins with collecting an initial video frame in which the
results are what was anticipated. The MLP classifier using moving objects are separated from the background.
the HOG descriptor pro- duced the most favorable
outcomes, with a 0.9137 accuracy. Processing merely the first few frames allow you to
take the necessary steps to process the video. To establish 6
The tracking algorithm used was the Lucas-Kanade the Gaussian mixture model, the foreground detector
tracker algorithm [12]. The tracking of objects predicts when requires a fixed number of video frames The foreground
a car andan individual will crash. The technology will sound segmentation procedure is rarely perfect and frequently
a warning if a car and a pedestrian cross the crossing at the contains unwanted noise. Following that, we locate the
same time. Given that a pedestrian may alter their course bounding boxes of each connected component that
after the system provides a warning, there is a high risk of corresponds to a moving vehicle.
false positives with this technology.
The next step is to categorize the moving vehicle
According to Dr. A. Radhika [13], The convolutional extracted in the previous section. A vehicle can be divided
neural network-based method detects helmet wear by into two types: two wheelers and four wheelers. Because we
using riders of motorcycles and two-wheelers. To determine wish to identify the helmet, we are only intrigued in two
if bike riders are wearing safety helmets, the model uses wheelers. Only if a two-wheeler is identified does the
SSD Mobile Net and the YOLO algorithm. Three system proceed. Otherwise, it discards this vehicle and
portions are built from a collection of photos with various searches for new ones, and the cycle continues.
helmets in order to train and test the model. TensorFlow is a
framework used for training the model. A helmet detection We obtained the training data required for vehicle clas-
model is produced once the detection model’s mean average sification on our own. We photographed numerous autos in
precision (mAP) is stableduring training and testing. various positions. we were collected for both the two-
wheeler and four-wheeler groups. When there are an
equal number of training photos from both classes, the

IJISRT24MAY156 www.ijisrt.com 3
Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

problem of classimbalance is eliminated, and the classifier In summary, the Faster R-CNN model, a deep learning
performs better. approach used in the proposed methodology, is used to
manage the critical issue of the helmet violation detection in
The training images feature a car encircled by real-time surveillance films. The system’s goal is to increase
intriguing features including buildings, walkways, trees, and the helmet compliance among motorcyclists in order to
other loud objects. The images show a car as it would often promote road safety. The evaluation phase will shed light on
be seen on the road. It is convenient to employ synthetic the system’s efficacy and possible effects in practical
images since they enable the creation of a large variety of situations.
training samples through image augmentation. A separate
set of images is used to evaluate the classifier. This dataset
is adequate to train and evaluate the efficiency of different
deep learning techniques in order to assess the model’s
viability, even though it is not the most exemplary of
moving objects in the real world. The images underwent a
grayscale conversion. Raw pixel values were input into the
classifier.

To figure out if the person driving is using a helmet,

we utilize the same technique used to determine the kind of
vehicle. A helmet detector is trained using edited photos of
two-wheeler photos with the rider’s helmet region in the
center. We were nevertheless managed to keep the class
balance by employing this strategy, as evidenced by the
equal number of photos including cyclists wearing helmets
and those without. To find the best machine learning classifier
for this assignment, we experimented with a number of
them.

IV. METHODOLOGY

The Automatic Helmet Violation Detection system

proposed in this research involves several key steps,
beginning with the collection of diverse video surveillance
data capturing various traffic scenarios with motorcycles.
The dataset undergoes annotation, where regions of interest
(ROIs) are labeled to identify instances of helmeted and
non-helmeted motorcy- clists. Pre-processing steps include
frame extraction and data augmentation to create a robust
and representative dataset for training and testing.

Because of its efficacy in object detection tasks, the

Faster R-CNN structure is selected as the main model for
helmet recognition. To take use of universal object detection
features, transfer learning is used by initializing the model
with pre- trained weights on a sizable dataset such as
COCO. Using the annotated dataset, the Region Proposal Fig 1 Flow Chart
Network (RPN) is trained, and the Faster R-CNN model is
refined. Stochastic gradient descent (SGD) is one of the  COCO
optimization strategies used to minimize the loss function The COCO (Common Objects in Context) dataset
and improve the accuracy of the model. serves as a valuable resource for training the Automatic
Helmet Violation Detection system. This dataset, widely
A approval set is used as part of the model evaluation used in computer vision research, contains a diverse array of
process to track performance during training and avoid over- images with detailed annotations, making it suitable for
fitting. Performance metrics like precision, recall, accuracy, pre-training deep learning models. By leveraging COCO,
and F1-score are used to assess how well the model detects the Faster R- CNN model is initialized with pre-trained
helmets. As part of the experimental results, real-time weights, enabling the system to learn generalized features
film is used to assess the trained model’s accuracy in for object detection. The richness of object categories and
recognizing helmet violations through live surveillance contextual information in COCO enhances the the model’s
testing. To assess the model’s performance under capacity to identify and cate- gorize various items,
demanding scenarios such as poor resolution, unfavorable contributing to the system’s effectiveness in identifying
weather, and occlusion, robust- ness testing is carried out. motorcycle helmets in real-time surveillancevideos.

IJISRT24MAY156 www.ijisrt.com 4
Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

 Faster-RCNN learning and machine learning models. TensorFlow is a

Faster R-CNN (Region-based Convolutional Neural distributed computing framework that leverages compu-
Net- work) represents a groundbreaking advancement in the tational graphs and can be used for a variety of purposes,
realm of object detection within computer vision. One of its such as recommendation systems, speech recognition,
key in- novations lies in the integration of a Region Proposal picture recognition, and natural language processing.
Network (RPN), which significantly enhances efficiency by TensorFlow is a popular tool among data scientists,
proposing candidate regions likely to contain objects. This engineers, and academics because it provides great per-
streamlining of the detection process mitigates the speed formance, flexibility, and a vast ecosystem of tools and
limitations that plagued earlier models, thereby enabling resources.
Faster R-CNN to operate swiftly and accurately even in
complex visual en- vironments. The model’s operation V. IMPLEMENTATION AND RESULTS
unfolds in two distinct stages: firstly, the RPN generates
these candidate regions, followed by a subsequent stage that The proposed Automatic Helmet Violation Detection
handles classification and refinement of these proposals. system leverages deep learning methods, particularly the
Through this two-stage ap- proach, Faster R-CNN achieves Faster RCNN model, to address challenges in real-time
a remarkable balance between precision and speed, rendering surveillance film, specifically focusing on the identification
it particularly well-suited for real-time applications where of motorcyclists wearing helmets. The issues faced by
rapid detection is imperative. This versatility makes it an existing models, such as limited resolution, inadequate
invaluable tool in numerous domains, including lighting, adverse weather, and occlusion, are targeted by a
surveillance, autonomous driving, and industrial inspection. unique approach. The method- ology involves training the
In the specific context of the Automatic Helmet Violation Region Proposal Network (RPN) first, using the RPN
Detection system, Faster R-CNN proves its mettle by weights to then train the Faster R-CNN model. The
efficiently detecting and classifying motorcycle helmets objective is to enhance the accuracy of helmet detection in
within surveillance videos. This application underscores the live surveillance footage.
model’s ability to overcome various challenges, such as
limited resolution, adverse weather conditions, and In terms of experimental results, the system exhibits
occlusions, thereby showcasing its robustness in practical promis- ing outcomes, achieving a commendable accuracy
scenarios. rate in iden- tifying motorcycle helmets in real-time
surveillance scenarios. This success underscores the
 Libraries capability of machine learning methodologies in automatic
helmet detection for motorcyclists, emphasizing the
 Pandas is a powerful Python library designed for effi- effectiveness of the proposed strategy in overcoming
cient data manipulation and analysis. It is particularly challenges faced by current models.
well-suited for handling large datasets, and it incorpo-
rates the capabilities of NumPy. Pandas is instrumental The work emphasizes the critical role of technology in
in performing extensive calculations and operations on addressing the rising issue of motorcycle accidents,
structured data. particu- larly due to non-compliance with helmet laws.
 NumPy is a versatile array processing library that en- With a focus on countries like India, where motorcycle
hances the performance of complex array objects in usage is high, the implementation of a computer vision-
Python. It is an essential package for scientific com- based automated system for helmet detection is seen as a
puting, providing support for large, multidimensional cost-effective solution. The system’s integration with
arrays and matrices. NumPy is crucial for high-level existing video surveillance infras- tructure is highlighted,
mathematical functions and operations on these arrays. emphasizing its potential to alleviatethe workload of traffic
on these clusters. police and improve overall road safety. The primary goal of
 Matplotlib is a 2D plotting library for Python, enabling the study is to prevent significant injuries resulting from
the creation of high-quality visualizations for various motorcycle accidents by deploying computer vision-based
print and web formats. With Matplotlib, users can gen- automatic helmet recognition. The proposed solution aims
erate a wide range of statistical plots, including line to identify riders violating helmet laws in real-time and
charts, histograms, power spectra, bar plots, contour alert authorities for appropriate action. This technology is
plots, scatterplots, and more. positioned as a vital tool in promoting compliance with
 A complete Python machine learning library is called helmet laws and contributing to overall road safety.
Scikit-learn. It includes numerous techniques for group-
ing, regression, classification, and other purposes. Nu- In summary, the increasing number of motorcycle
merous statistical and machine learning models, includ- accidents and head injuries among riders underscores the
ing DBSCAN, k-means clustering, decision trees, and importance of implementing measures to encourage helmet
support vector machines, are supported by Scikit-learn. usage. The proposed solution, utilizing computer vision-
It combines with SciPy and NumPy with ease. based automated helmet detection technology, is positioned
 Google created the open-source machine learning frame- as an effective approach, especially in regions with high
work TensorFlow. It offers a strong and adaptable motorcycle usage. The primary goal remains the prevention
platform for building, honing, and implementing deep of significant injuries caused by traffic accidents by
promoting compliance with helmet laws through automatic

IJISRT24MAY156 www.ijisrt.com 5
Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

helmet detection systems. VI. CONCLUSION

The Faster R-CNN architecture is chosen for helmet In conclusion, the Automatic Helmet Violation
de- tection, with transfer learning using pre-trained weights Detection system offers a highly effective solution to
from the COCO dataset. Performance indicators like enhance road safety, particularly in regions with high
accuracy, pre- cision, recall, and F1-score are used in the motorcycle usage. The proposed methodology, leveraging
assessment. Also, live surveillance testing is conducted to deep learning techniques and the Faster R-CNN model,
assess the model’s accuracy in identifying helmet violations demonstrates promising results, achieving a high accuracy
under challenging conditions. rate in real-time surveillance scenar- ios. This research
addresses the pressing issue of motorcycle accidents by
The inclusion of the COCO dataset is highlighted as a providing a cost-effective, automated system to enforce and
valuable resource for training the system, providing a promote helmet laws.
diverse set of images with detailed annotations. The Faster
R-CNN model’s effectiveness in object recognition tasks, The significance of the project lies in its potential to
coupled with the Region Proposal Network, is emphasized reduce the workload of traffic police through a proactive
for real- time applications, particularly in the context of the approach to helmet compliance. The real-time detection and
Automatic Helmet Violation Detection system. immediate alerting of authorities contribute to improving
overall road safety and preventing significant injuries
The implementation of the proposed system involves resulting from traffic accidents. The study provides a
the utilization of key libraries such as Pandas, NumPy, comprehensive overview of helmet detection
Matplotlib, Scikit-learn, and TensorFlow. These libraries methodologies, showcasing the evolution from traditional
contribute to data manipulation, analysis, visualization, and computer vision to advanced deep learning models.
machine learning functionalities. TensorFlow, in particular,
serves as the core machine learning framework for building, By incorporating the COCO dataset and utilizing key
training, and de- ploying the Faster R-CNN model. libraries such as Pandas, NumPy, Matplotlib, Scikit-learn,
and TensorFlow, the research demonstrates a well-
In conclusion, the research presents a comprehensive rounded implementation strategy. The proposed system’s
ap- proach to addressing the critical issue of helmet reliance on TensorFlow underscores its scalability and
violation de- tection among motorcyclists using advanced applicability. In summary, the Automatic Helmet Violation
computer vision techniques. The combination of deep Detection system emerges as a promising technological
learning methodologies, model selection, and rigorous intervention, offering a viable solution to encourage helmet
evaluation demonstrates the potential impact of the compliance and reduce head injuries among motorcyclists.
proposed system on improving road safety by promoting
helmet compliance. REFERENCES

[1]. Kunal Dahiya, Dinesh Singh, C. Krishna Mohan,

”Automatic Detection of Bike-riders without Helmet
using Surveillance Videos in Real-time”, Visual
Learning and Intelligence Group (VIGIL),2018
[2]. Chiu C-C, Min-Yu K, Chen H-T (2007) Motorcycle
detection and tracking system with occlusion
segmentation. In: Eighth International workshop on
image analysis for multimedia interactive services.
USA
[3]. Leelasantitham A, Wongseree W (2008) Detection
and classification of moving Thai vehicles based on
traffic engineering knowledge. In: International
Fig 2 Helmet Detection-1 conference on intelligent transportation system
telecom- munications
[4]. M. Swapna,Tahniyath Wajeeh,Shaziya Jabeen,”A
Hybrid Approach for Helmet Detection for Riders
Safety using Image Processing, Machine Learning,
Artificial Intelligence”,International Journal of
Computer Ap- plications,volume 182 – No. 37,
January 2019
[5]. Messelodi S, Modena C, Zanin M (2005) A computer
vision system for the detection and classification of
vehicles at urban road intersections.

Fig 3 Helmet Detection-2

IJISRT24MAY156 www.ijisrt.com 6
Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY156

[6]. Rohith C A, Shilpa A Nair, Parvathi Sanil Nair,

Sneha Alphonsa, Nithin Prince John,”An Efficient
Helmet Detection for MVD using Deep
learning”,@2019 IEEE
[7]. Prity Kumari, Pawan Kumar,”Helmet Detection on
Two-Wheeler Riders Using Machine
Learning”,IRJET, 03 Mar 2022
[8]. Romuere, Silva ;Kelson, Aires ; Thiago, Santos ;
Kalyf, Abdala ; Rodrigo, Veras ; and André,
Soares 2013. Automatic detection of motorcyclists
without helmet. 2013 XXXIX Latin America
Computing Conference (CLEI).
[9]. Apoorva Saumya, Gayathri.V, Dr. Venkateswaran K,
Sarthak Kale, Dr.Sridhar N “Machine Learning based
Surveillance System for De- tection of Bike Riders
without Helmet and Triple Rides”,IEEE,2020
[10]. Narong Boonsirisumpun, Wichai Puarungroj, and
Phonratichi Wairotchanaphuttha , ”Automatic
Detector for Bikers with no Helmet using Deep
Learning”, ©2018 IEEE
[11]. Romuere Silva, Kelson Aires and Rodrigo Veras,”
Helmet Detection on Motorcyclists Using Image
Descriptors and Classifiers”,IEEE,2014.
[12]. Lucas BD, Kanade T (1981) An iterative image
registration technique with an application to stereo
vision. In: Proceedings of the 7th inter- national joint
conference on artificial intelligence. Morgan
Kaufmann Publishers Inc, San Francisco, pp674–
679
[13]. Dr. A. Radhika, Associate Professor “Motorcycle
Safety Helmet Detec- tion for the Two-Wheeler
Motorcyclist”, International Research Journal of
Engineering and Technology (IRJET), Volume: 08
Issue: 08 — Aug 2021
[14]. C. Vishnu, Dinesh Singh, C. Krishna Mohan and
Sobhan Babu” De- tection of Motorcyclists without
Helmet in Videos using Convolutional Neural
Network”, ©2017 IEEE
[15]. Mistry, J., Misraa, A. K., Agarwal, M., Vyas, A.,
Chudasama, V. M., Upla, K. P. (2017). An automatic
detection of helmeted and non- helmeted
motorcyclist with license plate extraction using
convolutional neural network. 2017 Seventh
International Conference on Image Pro- cessing
Theory, Tools and Applications (IPTA).

IJISRT24MAY156 www.ijisrt.com 7