홈 › Papers

Research on Smoking Behavior Detection System Based on Deep Learning

As we all know,smoking endangers the health of smokers,and the harm of second-hand smoke to the health of people around us can not be ignored;in addition, improper smoking can sometimes cause many safety accidents,such as fire or explosion, and bring huge property losses to the society.Therefore,strengthening the supervision of smoking prohibition in public places has become a growing concern of all sectors of society.Intelligent smoking behavior detection system has become a very urgent demand,and the core of this kind of system is how to design an algorithm to detect smoking behavior quickly and accurately.According to the existing research status of smoking behavior detection,this dissertation proposes a smoking behavior detection method combined with face detection.The method is tested on the self-made smoking behavior data set,and its accuracy is 94.78%,and its detection speed is 26.76FPS, which basically meets the requirements of accuracy and real-time of smoking behavior detection.The main research contents of this dissertation are as follows: (1)In view of the misjudgment of smoking behavior that may occur in the direct detection of cigarettes,a smoking behavior detection method combining face detection and cigarette detection is proposed.The face detection model and target detection model are used to detect faces and cigarettes respectively,and whether there is smoking behavior is judged according to the position relationship between face detection frame and cigarette detection frame. (2)Pyramid feature fusion is carried out on the existing face detection model FaceBoxes,which enhances the ability of the model to detect multi-scale faces,and adjusts the structure of rapid digestion convolution layer to retain more face feature information. (3)Several powerful deep learning target detection algorithms are studied,and their performance on the self-made smoking behavior data set is tested.According to the experimental results,YOLOv4 with excellent performance is selected for subsequent improvement to adapt to the detection of targets such as cigarettes. (4)In order to improve the detection ability of YOLOv4 for the smaller cigarette target in the image,this dissertation improves the neck of YOLOv4,improves the accuracy of the model,and uses PP-LCNet to replace the original backbone network CSPDarknet53,which speeds up the reasoning speed. (5)According to the existing web application development technology and the smoking behavior detection method in this dissertation,a simple online smoking behavior detection system is designed,which can realize the interaction between users and the system,and provide the possibility for better applying the smoking behavior detection algorithm to the actual society.

Code (0)

등록된 구현이 없습니다.

Tasks

Methods 이 논문이 사용한 방법론

Similar Papers 제목 키워드 기반