DefinePK

Abstract

Background: In 2023, over 50% of road fatalities in India were linked to motorcycle riders not wearing helmets. The inadequacy of current road safety systems in detecting such violations in real-time necessitates the development of more advanced, automated solutions.
Objectives: This study aims to develop a real-time system that accurately detects helmet usage and recognizes motorcycle license plates using state-of-the-art machine learning techniques, thereby reducing road fatalities and improving traffic law enforcement.
Materials and Methods: The system integrates Convolutional Neural Networks (CNN) for helmet detection and the YOLOv3 (You Only Look Once, version 3) model for motorcycle and license plate recognition. The input video is resized and pre-processed before being passed to the YOLOv3 model, which identifies and localizes motorcycles and license plates using bounding boxes and confidence scores. Non-maximum suppression is applied to refine the detections. For each motorcycle detected, a region of interest (ROI) is cropped and analysed by the CNN model to determine helmet usage, with classification output as "helmet" or "no helmet." The annotated video is stored for further evaluation.
Results: The system effectively detects and labels motorcycles and helmet compliance in real-time. Using video inputs with a resolution of 888 × 500 pixels, the model achieved a mean average precision (mAP) of 54.64%. The integration of CNN and YOLOv3 enhanced detection accuracy and system responsiveness.
Conclusion: The proposed system demonstrates the potential of deep learning-based approaches to significantly improve road safety by enabling real-time helmet detection and license plate recognition. This integrated solution can aid authorities in enforcing helmet laws and reducing motorcycle-related fatalities.