<![CDATA[Smoking is a serious health problem that increases the risk of chronic diseases and reduces the quality of life of the community. In the campus environment, smoke-free zone (SFZ) policies are often difficult to enforce due to limited supervision and low compliance among the academic community. To address this challenge, this study developed a smart surveillance system based on IP cameras and computer vision technology using the YOLO-pose algorithm. The research aim is to produce a system capable of automatically detecting smoking activity, thereby supporting the enforcement of SFZ policies in educational settings. The research methods include literature study, problem analysis, dataset collection through annotation using the Pose Annotator Tool, detection model training using YOLO11m-pose, and performance evaluation using standard metrics. The dataset was constructed by adding body keypoint annotations, including the characteristic hand-to-mouth gesture pattern associated with smoking activity. The model was trained for 300 epochs and tested using GPU and CPU hardware to measure inference speed. The results showed that the model achieved very high performance, with precision and recall values of 1.000 and mAP50 of 0.995. The system can operate in real-time at an average speed of 45–60 FPS on a mid-range GPU and continues to run on a CPU at 8–12 FPS. This study concludes that the integration of custom pose annotations with the YOLO-pose architecture can improve the effectiveness of smoking activity detection compared to static object-based approaches]]>
Copyrights © 2026
