Multiple Machine Learning Algorithms for Human Smoking Behavior Detection
Chenxin Cui, Ruofeng Xu · 2022 International Conference on Machine Learning and Intelligent Systems Engineering (MLISE) · 2022
Due to the harm and accidents caused by smoking behavior, people have to spend a great number of efforts to detect those behavior in many public places by people's eyes. This way of detection is sometimes inaccurate and exhausting. In this paper, three models were proposed to find smoking behavior automatically. The dataset in this paper has three classes: smoking, calling and normal (neither smoking nor calling). Smoking behavior detection is different from cigarette detection or smog detection since it needs us to find a smoker first. Therefore face detection was used to determine if there is a person in the picture. Then a screenshot will be made on the picture, and this screenshot is supposed to include the person's face and the area around the face. After that this screenshot will be turned to grayscale and be resized to 64× 64. Data augmentation is used as well to make proposed model more robust. There are 3 different models used in this study: support vector machine (SVM), Random Forest and convolutional neural network (CNN). Compared to SVM and Random Forest, in CNN model the study got the best performance with an accuracy of 94.59% in testing dataset which consists of 522 images. Experiments show the high efficiency of our method, and smoking behavior can be detected accurately.