Deep Convolutional Neural Network for Detection of Cigarette Smokers in Public Places: A Low Sample Size Training Data Approach

Cigarette smoking is not only known to be harmful to the smokers' health but also the people around them and despite the efforts made by the government such as creating policies for controlling tobacco use and increasing tax rates, second hand and third-hand smoke that may harm the health of the people in public still occur. A solution for detecting smoking violators in public places is presented. This will be beneficial to the public as each smoker can affect many individuals going to their work or just having fun outside of their homes that do not smoke but still suffer from second-hand smoke or even third-hand smoke. To create a smoking detection, the YOLOv3 model was used to develop a not smoking and smoking detection model. The detection model that was created for this paper has a training and validation accuracy of 96.2% with an mAP value of 91.36%.