A Smoking Behavior Detection Algorithm Based on YOLOv5 and Key Points of Human Skeleton

As people realize the harm of smoking in public places, the problem of how to improve the detection efficiency of smoking behavior in surveillance video remains to be solved. In recent years, the algorithm of pose estimation based on the key points of human bones has become more and more mature. In this paper, an algorithm for smoking behavior detection based on the relationship between the position information of the key points of human bones and smoking behavior is proposed. The algorithm first uses AlphaPose to detect the key point information of human bones, and determines whether the behavior is smoking by calculating the distance ratio (SBR) between the wrist and the nose and the wrist and the eye on the same side. YOLOv5 was used to detect whether there were cigarette butts in the video. Weight fusion of cigarette probability detected by SBR and YOLOv5 was performed to determine whether there was smoking behavior. The experimental results show that the proposed method can effectively detect smoking behavior with an accuracy of 93.3%. Finally, the abnormal behavior recognition system in the library environment is developed through Python3.8 and PyQt development languages, and the two algorithms proposed in this paper are integrated into the system to realize the visualization of the two abnormal behavior recognition results, which proves the effectiveness of the abnormal behavior recognition algorithm proposed in this paper, and also shows that the system has certain practical application value.