Helmet Detection Using Machine learning and Deep learning

Techniques
Ganesh Sakat1, Yash Ruikar2, Nikunj Sakhare3, Prathamesh
Solunke4 ,Tushar A Rane 5

Department of Information Technology, SCTR’s

Pune Institute of Computer Technology, Pune, Maharashtra, India

Corresponding author(s). E-mail(s):

[email protected];[email protected]; [email protected];
[email protected]; [email protected]

ABSTRACT
Motorcycle accidents are a growing concern in developing countries, with one of the leading causes of
fatalities being the lack of helmet usage by riders. Traditional methods, such as manual monitoring by traffic
police or reviewing CCTV footage, are labor-intensive and require significant human effort. This paper
proposes an automated system to address this issue by utilizing CCTV footage to detect motorcyclists not
wearing helmets and retrieving their license plate information. The system first differentiates between
motorcycles and other objects in motion. For motorcyclists, it identifies whether they are wearing
helmets or not. If a rider is detected without a helmet, the system employs an Optical Character
Recognition (OCR) algorithm to extract the license plate number. The proposed system offers a more
efficient approach to monitoring helmet compliance and enhancing road safety, minimizing the need for
human intervention. Key technologies employed include the YOLO algorithm for object detection and
classification, and OCR for license plate extraction.

Keywords: Helmet detection, motorcycle safety, YOLO algorithm, object detection,

CCTV monitoring, optical character recognition (OCR), license plate recognition,
road safety, automated traffic monitoring, deep learning, image processing, non-
helmeted riders, computer vision.

INTRODUCTION

Motorcycle accidents pose a significant public safety issue, particularly in developing countries where traffic
regulations are poorly enforced, leading to high rates of fatalities and injuries among motorcyclists. A primary
factor contributing to these statistics is the low compliance with helmet-wearing laws, which are crucial for
reducing head injuries during accidents [1].
Traditional methods for ensuring helmet compliance, such as manual monitoring by traffic police and reviewing
CCTV footage, are labor-intensive and inefficient, often leading to errors due to human fatigue [2], [3]. To
tackle these challenges, we propose an automated helmet detection system that utilizes advanced computer
vision techniques to detect motorcyclists not wearing helmets in real time using CCTV footage [4].
Our system employs a two-stage approach. First, it utilizes the YOLO (You Only Look Once) algorithm, known
for its speed and accuracy, to distinguish motorcycles and detect helmet usage [5], [6]. If a motorcyclist without
a helmet is identified, the system then extracts the license plate number using Optical Character Recognition
(OCR) technology [7], [8]. This automated process enhances enforcement capabilities, allowing authorities to
issue fines or warnings without manual footage review [9].
The integration of such technology can significantly improve road safety and helmet compliance in developing
regions. By leveraging existing surveillance infrastructure, traffic authorities can efficiently monitor helmet usage,
thus reducing motorcycle-related fatalities and injuries [10], [11]. This paper explores the technical architecture of
the proposed system, detailing the YOLO algorithm’s application for object detection and OCR for license plate
recognition, alongside performance evaluations in various real-world scenarios. Through this research, we aim to
demonstrate the effectiveness of automated helmet detection as a means to enhance road safety [12], [13].

LITERATURE SURVEY

[1] Detection of Helmets on Motorcyclists (Romuere R.V., Silva, Kelson R.T. Aires, Rodrigo de M.S. Veras,
2017 This paper introduces a helmet detection system utilizing image processing and machine learning
techniques, including the Circular Hough Transform, Histogram of Oriented Gradients (HOG), and a MultiLayer
Perceptron (MLP) classifier. It highlights the importance of helmet use in reducing motorcycle accident
fatalities. With a dataset of 255 traffic images, the system achieved an accuracy of 91.37%.[2] Helmet Detection
and Number Plate Recognition (Tushar Kale, Alisha Dhavale, Aniket Randhavan, Mrs. V. S. Inamdar, May
2023) .This study proposes a fully automated system for detecting helmet violations and recognizing license
plates using real-time CCTV footage. It employs the YOLOv5 algorithm for object detection and Optical
Character Recognition (OCR) for license plate extraction, significantly enhancing efficiency and reducing the
need for human intervention.[3] A Review on Helmet and Number Plate Detection (Priyanshi Tripathia, Pragati
Singha, Mantsha Banoa, Komal Sharmaa, Abhishek Shahib, 2023) This comprehensive review discusses various
helmet and number plate detection methods, focusing on the YOLOv3 algorithm, OpenCV, and TensorFlow. It
emphasizes accurate data collection, model training, and potential applications in law enforcement and toll
collection to enhance road safety.
[4] Helmet and Number Plate Detection (Swapna Rani Kura, Suman Rathlavath, Bharath Simha Reddy,
Linganagari Sai Kumar Ravindra Golla, Ajay Vislavath, Navdeep Singh, May 2023)
This paper presents a helmet and number plate detection system using the SSD (Single Shot Multibox Detector)
algorithm, enabling real-time detection of motorcyclists not wearing helmets. The system aims to enhance road
safety and traffic management by automating the identification of traffic violations.[5] Real-time helmet
detection using deep learning (A. Sharma, N. Jindal, 2020).This paper presents a deep learning approach for
real-time helmet detection using CNNs, demonstrating high accuracy and processing speed, contributing to
enhanced road safety.

[6] Helmet Detection in Traffic Surveillance (R. Kumar, S. Gupta, 2021)

The authors propose an automated helmet detection system leveraging YOLOv4, achieving a detection accuracy
of over 90%. This system enhances the enforcement of helmet laws through efficient monitoring.[7] Multi-
Object Detection for Traffic Safety (M. Iqbal, T. Rashid, 2022) This study discusses the integration of multiple
object detection techniques for identifying helmet usage and vehicle type, focusing on the application of
advanced computer vision models for traffic safety.[8] Challenges in Helmet Detection Systems (H. Patel, R.
Sharma, 2022) The authors explore the challenges faced by helmet detection systems in urban environments,
proposing solutions to improve accuracy and reduce false positives in detection. [9] Automated Traffic
Surveillance Using AI (P. Sen,
A. Malhotra, 2021) This paper investigates the use of artificial intelligence for traffic surveillance, specifically
focusing on the detection of helmet use among motorcyclists and integrating with traffic management systems.
[10] Innovations in Road Safety Technology (K. Das, L. Verma, 2022) .This study reviews innovative
technologies in road safety, including helmet detection systems, highlighting the role of machine learning in
enhancing traffic law compliance.[11] Deep Learning Approaches for Helmet Detection (Z. Chen, X. Liu, 2023)
This paper discusses various deep learning frameworks for helmet detection, showcasing the effectiveness of
model optimization techniques to enhance detection performance in diverse condition
METHODOLOGY

The goal of this research is to develop an efficient system for detecting helmet violations using real-time
object detection models, namely YOLOv3 and YOLOv5, alongside Optical Character Recognition (OCR) for
vehicle identification through number plate recognition. This section presents a detailed description of the
methodology, breaking down the steps and processes involved.

1. Helmet Detection Using YOLOv3 and YOLOv5

YOLOv3 and YOLOv5 are state-of-the-art object detection models known for their high-speed processing and
accuracy in real-time scenarios. These models are crucial for detecting helmets, as they allow for swift
identification even in complex or cluttered backgrounds such as traffic scenes.

1.1 YOLOv3 Architecture and Functionality

YOLOv3 employs Darknet-53, which has 53 convolutional layers, each of which applies multiple filters to the
input image to extract features like edges, textures, and shapes. YOLOv3 is particularly useful for multi-scale
object detection, which is vital in traffic monitoring, where helmet sizes and shapes can vary across different
images and scenes.
YOLOv3 processes images in three scales—13x13, 26x26, and 52x52—allowing it to detect both large and
small objects. This is achieved through bounding box regression, where the model predicts the coordinates
and dimensions of an object within an image.
Key Formula for Bounding Box Calculation in YOLOv3:

1.2 YOLOv5 Architecture and Advancements

YOLOv5, a more recent model, offers significant improvements in accuracy and speed over YOLOv3. It
introduces Cross-Stage Partial Networks (CSPNet), which optimize memory usage and improve inference time.
YOLOv5 also uses Path Aggregation Networks (PANet) to enhance feature pyramid extraction, which improves
the detection of smaller objects like helmets, particularly in dense traffic situations.

YOLOv5’s superior GIoU (Generalized Intersection over Union) and CIoU (Complete IoU) losses allow it to
improve bounding box accuracy during object detection. The CIoU formula is designed to enhance detection
performance by considering the overlapping area, center point, and aspect ratio, making it particularly
effective for real-time applications.
1.3 Comparative Performance of YOLOv3 and YOLOv5
The performance of YOLOv3 and YOLOv5 is compared based on key metrics such as precision, recall, F1
score, and frames per second (FPS), which are critical for real-time applications. A detailed comparison is shown
in Table 1, based on our experimental results.

From the table, YOLOv5 exhibits better precision and recall, demonstrating its ability to detect more helmets
accurately and faster compared to YOLOv3. This makes YOLOv5 a more suitable model for real-time helmet
detection in crowded traffic scenarios.

2. Preprocessing Steps and Data Augmentation

To ensure the models perform well in diverse environmental conditions, the dataset is preprocessed and
augmented to create more robust and adaptable detection models. Key preprocessing steps include:

1. Resizing: All input images are resized to 416x416 (for YOLOv3) or 640x640 (for YOLOv5), which ensures
consistency and compatibility with the input layers of the models.
2. Normalization: Image pixel values are normalized to fall between 0 and 1, reducing the variance and making
model training more stable.
3. Augmentation: Random flips, rotations, zooms, and brightness adjustments are applied to the dataset to
simulate real-world conditions, enhancing model generalization.

Table 2 shows the augmentation techniques applied to the dataset.

3. Number Plate Detection Using OCR
Once a helmet violation is detected, the system employs Optical Character Recognition (OCR) to extract
the number plate from the image. The process involves several steps:

1. Segmentation: The number plate is segmented from the image using contour detection techniques.
2. Character Recognition: Using Tesseract OCR, characters from the segmented number plate are recognized
and converted into alphanumeric text.

The effectiveness of the OCR system is quantified using the Recognition Rate (RR) formula:

OCR performance is dependent on the image quality, lighting conditions, and preprocessing steps like
thresholding and noise reduction. Enhancing the quality of input images significantly improves the character
recognition accuracy.

4. System Integration and Workflow

The complete system integrates helmet detection and number plate recognition into a single framework that
operates in real-time. The workflow is illustrated in Figure 1, where each step is represented sequentially:
1. Input: Capture images from a surveillance feed.
2. Helmet Detection: Apply YOLOv3 or YOLOv5 for helmet detection.
3. Violation Detection: If no helmet is detected, proceed to OCR.
4. Number Plate Recognition: Extract and recognize the vehicle's number plate using OCR.
5. Data Logging: Store the violation record, including the number plate and timestamp.
FLOWCHART
Conclusion

This project demonstrates the significant potential of advanced technologies, including

Convolutional Neural Networks (CNNs), YOLO algorithms, and Optical Character Recognition
(OCR), in automating helmet and number plate detection to enhance road safety. By utilizing real-
time CCTV footage, the proposed system effectively identifies motorcyclists who are not wearing
helmets and retrieves their license plate information for further enforcement actions. This integration
of machine learning and image processing techniques enables high-accuracy real-time detection,
significantly reducing the reliance on manual monitoring by traffic authorities.The framework not
only promotes compliance with helmet laws but also offers an adaptable and scalable solution
suitable for various environments, particularly in developing countries where motorcycle accidents are a
growing concern. The system's ability to operate efficiently in diverse conditions makes it a vital tool
for improving overall road safety.

Future Scope

Looking ahead, several advancements can be explored to enhance the system's performance further. The
incorporation of edge computing technologies could optimize processing speed and reduce latency, enabling
real-time applications even in areas with limited bandwidth. Additionally, investigating more sophisticated
machine learning models may improve detection accuracy, allowing for better differentiation between
motorcyclists wearing helmets and those who are not.Moreover, the future implementation of this technology
could extend beyond helmet detection to include other traffic regulations, creating a comprehensive traffic
management system. Integrating the system into smart city initiatives could facilitate seamless data sharing
among traffic authorities, enhancing overall traffic monitoring and law enforcement efforts.Overall, this
technology presents a promising pathway for mitigating motorcycle accidents, strengthening law enforcement
actions, and ultimately saving lives. By embracing these innovations, we can pave the way for safer roadways
and foster a culture of compliance with traffic regulations.
REFERENCES

[1] N. Sharma, R. Sharma, and N. Jindal, "Machine learning and deep learning applications—A vision,"
Global Transitions Proc., vol. 2, no. 1, pp. 24–28, Jun. 2024, doi: 10.1016/j.gltp.2024.01.004.

[2] L. Yang, L. Cai, and S. Gu, "Detection on wearing behavior of safety helmet based on machine
learning method," J. Saf. Sci. Technol., vol. 15, no. 10, pp. 152–157, 2019.

[3] K. Zhang, Y. Wu, J. Wang, Y. Wang, and Q. Wang, "Semantic context-aware network for multiscale
object detection in remote sensing images," IEEE Geosci. Remote Sens. Lett., vol. 19, pp. 1–5, 2022.

[4] L. Wang, J. Duan, and L. Xin, "YOLOv5 helmet wear detection method with introduction of attention
mechanism," Comput. Eng. Appl., vol. 58, no. 9, pp. 303–312, 2022.

[5] Z. Chen, F. Zhang, H. Liu, L. Wang, Q. Zhang, and L. Guo, "Real-time detection algorithm of helmet
and reflective vest based on improved YOLOv5," J. Real-Time Image Process., vol. 20, no. 1, p. 4, Feb.
2023.

[6] Y.-F. Zhang, W. Ren, Z. Zhang, Z. Jia, L. Wang, and T. Tan, "Focal and efficient IOU loss for
accurate bounding box regression," Neurocomputing, vol. 506, pp. 146–157, Sep. 2022.

[7] L. Yang, et al., "Deep learning techniques for road safety: A comprehensive study," J. Saf. Technol.,
vol. 15, pp. 140–149, 2020.

[8] Q. Zhou, Y. Li, and B. Zheng, "Real-time helmet detection using YOLOv4 in complex environments,"
Proc. Comput. Vision, vol. 45, pp. 200–205, 2021.

[9] X. Liu, R. Feng, "OCR-based detection in surveillance footage: Combining object detection and text
recognition," Int. J. Machine Vision, vol. 18, pp. 50-58, 2021.

[10] Enhancing IoT Device Classification with Hybrid Stacked Ensembles of Machine Learning
Classifiers GS Pise, SD Babar, PN Mahalle
International Journal of Intelligent Systems and Applications in Engineering

[11] L. Chen, "Improving helmet detection accuracy using deep learning algorithms," Safety Autom., vol.
17, pp. 90–98, 2022.

[12] A. Kumar and S. Jain, "An Enhanced Helmet Detection System Using Deep Learning Techniques,"
International Journal of Computer Applications, vol. 182, no. 39, pp. 9–15, 2023.

[13] M. Patel and V. Desai, "Application of YOLOv3 for Helmet Detection in Real-Time Surveillance,"
Journal of Electrical Engineering and Automation, vol. 5, no. 2, pp. 50–58, 2023.

[14] T. Singh, A. K. Sahu, and N. Gupta, "Improving Object Detection in Complex Scenes with YOLOv3,"
Journal of Computer and Communications, vol. 11, no. 1, pp. 13–21, 2023.

[15] J. Brown, K. Roberts, and L. Smith, "A Comparative Study of Deep Learning Techniques for Traffic
Monitoring," International Journal of Advanced Computer Science and Applications, vol. 14, no. 5, pp.
145–150, 2023.

[16] Investigating Recent Advances In Coded Diffraction Patterns using Deep

Learning AA Pise, S Singh, K Hemachandran, S Gadilkar, Z Benisemeni, ...
Full Length Article 7 (1), 62-2-71
[17] R. Gupta and A. Sharma, "Machine learning for traffic monitoring and helmet detection," IEEE
Trans. Veh. Technol., vol. 69, pp. 157–167, 2020.

[18] Utilizing Asymmetric Cryptography and Advanced Hashing Algorithms for Securing
Communication Channels in IoT Networks Against Cyber Espionage.
AA Pise, S Singh, S Gadilkar, ZB Esther, GS Pise, J Imuede
Journal of Cybersecurity & Information Management 13 (1)

[19] An Improved K-Means Clustering Process Solicitation for Mine Blood Donors
Information. AA Pise, GS Pise, S Singh, S Gadilkar, J Imuede, S Shinde
Journal of Cybersecurity & Information Management 13 (1)