Smoking behavior detection based on improved YOLOv7 object association

To address the security risks caused by unauthorized individuals smoking in public places, a method for detecting smoking behavior based on an improved YOLOv7 target association is proposed. First, the improved YOLOv7 algorithm is used to detect the people and cigarette targets in the image, and then the distance relationship between the center coordinates of the cigarette target detection frame and the person target detection box is calculated, and the correlation between the two is located to determine whether there is smoking behavior. The improved YOLOv7 algorithm introduces the CBAM attention module to prioritize key areas and diminish background interference. Simultaneously, the SIOU loss function is employed to bolster the precision of the regression outcomes. Experiments conducted on the test dataset reveal that the proposed approach achieves an accuracy of 91.35% and a detection speed of 78 FPS, achieving real-time accurate detection effects.