Smoke Sentry: Employing YOLO-based Deep Learning for Sophisticated Real-Time Smoking (Cigarette smoking) Detection in Surveillance Networks

Cigarette smoking in public places poses a real health risk and hence, requires effective supervision. In this study, we present a new method of identifying cigarette smoking behavior through the use of cigarette object detection with an improved YOLOv9 model. Our system will record and analyze the video frames grabbed from CCTV cameras to check for the presence of cigarettes in the scene and if smoking behavior is observed, then an automatic email containing the recorded video footage is sent. The main goal of this work is to investigate the effectiveness of the YOLOv9 model, which is developed in this study, concerning the YOLOv8 model in cigarette object detection in terms of accuracy, speed, and stability. Consumers of YOLOv9 have benefited from enhancements of object detection precision in difficult environmental conditions, including different lighting. As seen in the extensive experiments on different datasets, YOLOv9 performs better than others in detecting the cigarette object and avoiding interference from other objects in the scene. This paper introduces various changes to the architectural and training aspects of the YOLOv9 model and evaluates the results comparing it with YOLOv8. The presented model shows higher sensitivity compared with traditional approaches and is appropriate for real-world applications in the protection of people’s lives and health.