Bài viết Kết hợp mô hình YOLOv4-tiny và LSTM để nâng cao độ chính xác của hệ thống cảnh báo người có hành vi hút thuốc lá nơi công cộng trình bày các phương pháp nhận biết hành vi hút thuốc lá; Mô tả cấu trúc mạng kết hợp YOLOv4-tiny + LSTM; Mô tả sơ đồ tổng quan hệ thống cảnh báo hành vi hút thuốc. | Transport and Communications Science Journal Vol 73 Issue 8 10 2022 785-797 Transport and Communications Science Journal COMBINING YOLOV4-TINY AND LSTM MODELS TO ENHANCE ACCURACY OF THE WARNING SYSTEM OF SMOKING BEHAVIOR IN PUBLIC PLACE Vo Thien Linh Division of electrical and electronics engineering Campus in Ho Chi Minh City University of Transport and Communications No. 450 451 Le Van Viet Street Tang Nhon Phu A Ward Thu Duc City Ho Chi Minh City Vietnam ARTICLE INFO TYPE Research Article Received 24 08 2022 Revised 27 09 2022 Accepted 14 10 2022 Published online 15 10 2022 https Corresponding author Email linhvt_ph@ Tel 84 907001184 Abstract. Tobacco smoke contains many chemicals which cause typical dangerous diseases such as lung cancer cardiovascular disease infertility and many other incurable diseases. Monitoring and warning smokers in public places are one of the important jobs in propaganda and prevention of the harmful effects of tobacco. A smoking monitoring system and smart warnings have been researched and implemented by several groups so far. The objective of this study is to improve the previous system using Yolov4 and LSTM combination to give highly accurate prediction results. Instead of just using Yolov4 as a method of detecting people holding a cigarette this proposed method uses Yolov4 to extract features from the frames of the videos. This sequence of consecutive feature frames is fed into the LSTM network for prediction. To evaluate the method s performance the presented study performed on the group s own collected data set 5000 photos and 120 videos containing smoking behaviour. The results show that the proposed approach is successful in detecting human smoking actions on this dataset with higher accuracy in comparison with that in conventional method. Keywords Smoking behavior convolutional neural network Deep learning YOLOv4-tiny Long Short Term Memory. 2022 University of Transport and .
