A Deep Learning-Based CCTV System for Automatic Smoking Detection in Fire Exit Zones

This study proposes a deep learning-based system for real-time smoking detection in CCTV-monitored fire exit zones, addressing critical safety concerns. A comprehensive dataset of 8,124 images from 20 scenarios, along with 2,708 low-light samples, was developed to capture real-world complexity. We evaluated YOLOv8, YOLOv11, and YOLOv12 models, and introduced a custom architecture derived from YOLOv8, optimized for challenging surveillance environments. The proposed model achieved a recall of 78.90% and mAP@50 of 83.70%, outperforming existing models. Edge-device performance testing on the Jetson Xavier NX demonstrated real-time processing with 52–97 ms latency, validating suitability for time-sensitive applications. The system provides a scalable, automated solution for enforcing no-smoking policies and enhancing public safety compliance.

Recommended citation: Sami Sadat, Md. Irtiza Hossain, Junaid Ahmed Sifat, Suhail Haque Rafi, Md. Waseq Alauddin Alvi, Md. Khalilur Rhaman. (2026). "A Deep Learning-Based CCTV System for Automatic Smoking Detection in Fire Exit Zones." Preprint.
Download Paper