As awareness of smoking prohibition in public places increases, technology that automatically detects smoking behavior is particularly important. With the rapid development of deep learning and neural networks, the typical You Only Look Once version 5 (YOLOv5) algorithm can be used for prediction, the traditional algorithm performs well on the picture, but the detection on the video will have false detection, so the MediaPipe machine learning framework is introduced to improve the accuracy of video detection, and the relative distance calculation of the target’s hand, mouth, and the position of the cigarette detected by YOLOv5 can comprehensively judge whether the object smokes. Deploy an improved algorithm to call the camera or import the video for detection, and the items similar to cigarettes in the video can be eliminated, reducing the probability of false detection.