Real-Time Cigarette Smoking Detection Using YOLOv8 Convolutional Neural Network (CNN)

Abstract

Cigarette smoking has major health and environmental effects, especially in public places where second-hand smoke is a concern. Educational and higher education institution areas are among the 28 places categorized as areas effective January 1, 2025, following the implementation of the Smoking Products Control Act for Public Health 2024 (Act 852). To guarantee the efficient enforcement of the law, a real-time cigarette smoking detection system has been developed by using the YOLOv8 object detection model based on Convolutional Neural Networks (CNN). The system was built with Python and Streamlit, while the training and validation were done on a dataset of over 2,300 annotated photos tagged with ‘cigarette’, ‘smoking’, and ‘face’. The YOLOv8n (nano) model was chosen for its compact architecture and ability to identify small objects. Training was carried out in Google Colab, with further programming and testing in PyCharm. The trained model performed well, with a precision of 93.73%, recall of 90.36%, F1-score of 91.86%, and mean Average Precision (mAP@0.5) of 94.36%, indicating that it is suitable for real-world amazing showed promising performance deployment. The system includes a web interface for detection via camera or video upload, automatic logging in CSV format, PDF report generation and Telegram notifications. Overall, the study successfully shows the practical application of deep learning for scalable and accurate real-time cigarette smoking detection, thus promoting better enforcement of no-smoking laws and improving public health surveillance.