Detection of Smoking Behaviors Using Human Key Points and YOLOv8
W Zhang, Anton Louise P. De Ocampo, Rowell Hernandez · 2024
To effectively enhance smoking regulation and ensure strict enforcement of smoking bans, this paper proposes an intelligent smoking behavior detection method that combines human key point detection with the YOLOv8 algorithm for cigarette object detection. By leveraging deep learning for cigarette butt detection and incorporating human key point information, smoking actions are identified through calculations of distances, angles, and time cycles. The method first utilizes AlphaPose and RetinaFace to obtain human key point locations. Subsequently, based on these key points, distances between hands and mouths, angles among hands, elbows, and shoulders, and smoking time cycles are computed to establish smoking behavior rules. Finally, YOLOv8 is employed to detect the presence of cigarette butts in images, thereby judging the occurrence of smoking behaviors. Experimental results demonstrate that this method can detect smoking behaviors promptly and effectively.