International Research Journal of Engineering and Technology (IRJET)
e-ISSN:2395-0056
Volume: 12 Issue: 12 | Dec 2025
p-ISSN:2395-0072
www.irjet.net
Automatic Helmet Violation Detection Using Deep Learning Techniques Dr. Varsha M1, Niveditha MG 2, Priyadarshini P 3, Punyashree B S 4, Sahana M K 5 1Associate Professor, Artificial Intelligence and Machine Learning, Bapuji Institute of Engineering and Technology,
Davanagere, affiliated to VTU Belagavi, Karnataka, India.
2Bachelor of Engineering, Artificial Intelligence and Machine learning, Bapuji Institute of Engineering and
Technology, Karnataka, India
3Bachelor of Engineering, Artificial Intelligence and Machine learning, Bapuji Institute of Engineering and
Technology, Karnataka, India
4Bachelor of Engineering, Artificial Intelligence and Machine learning, Bapuji Institute of Engineering and
Technology, Karnataka, India
5Bachelor of Engineering, Artificial Intelligence and Machine learning, Bapuji Institute of Engineering and
Technology, Karnataka, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In India, most people prefer two-wheeler vehicles as their primary mode of transportation. It is a common form
in India. This usually leads to road accidents because of not wearing helmets. In developed cities, the digital system has been used to detect non-helmet riders. Often, this reduces the number of road accidents and the rate of death. But still, in India, there are many cities, even though all rules are followed. Some of the people are not wearing helmets and following the rules. To overcome this, helmet violations have been detected using CCTV surveillance cameras and machine learning models, including deep learning object detection models. This has reduced the rate of accidents and the rate of death. 44.5% of twowheelers, 74,897 people were killed due to this reason. In order to resolve this problem we have developed a model that is having accuracy 95.3, the model is trained on open source dataset, and used yolov8 for model training and integrated cub am for more accurate results even in low light conditions. Key Words: Helmet Violation Detection, Deep Learning, YOLOv8, CBAM (Convolutional Block Attention Module), Object Detection, Number Plate Recognition (OCR), CNN (Convolutional Neural Network), CCTV Surveillance, Echallah Automation, Twilit Messaging, India Traffic Safety, Performance Evaluation (Accuracy, Precision, Recall, map), Dataset Annotation and Augmentation, Smart City Deployment. 1. INTRODUCTION Traffic accidents are happening and leading to death, especially among two-wheelers, who are highly involved in this. The two-wheeler riders are getting injured due to not wearing a helmet, which causes death worldwide. In India, over the past 10 years, non-compliance has remained a significant issue, contributing to an increased number of road accidents. Approximately 30% of total deaths on roads were caused by helmet-violating riders. [1] To overcome this issue, digital informants have been undertaken in India. The use of technologies like CCTV cameras and artificial intelligence systems [2] automatically identifies and finds violations. In this paper, we are going to frame one solution for this. Nowadays, with the improved technologies, helmet violations are decreased using machine learning models like YOLOv8, OCR for Number plate Recognition, and Twilio for message alert. You Only Look Once is the most popular object detection model in machine learning and deep learning. [3] Over the past five years, the effectiveness of machine learning in CCTV has become more efficient. Various research papers and studies show that high. Accuracy of detecting violations using deep learning technologies, such as CNN, i.e., Computational Neural Network, typically ranges from over 90% to 98.56% in controlled experiments.[5] The impact of ML in this is enabling automated detection and the generation of each CNN and notification, which is faster than manual monitoring. The World Health Organization data indicates a reduction in the rate of death by 42% and severe head injuries by up to 69%. This system contributes to a reduction in the deaths or injuries of life. The development of real-time solutions can reduce the number of deaths and injuries.
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 255