Detection of Tobacco Use through Motion Analysis from Camera Images

Tobacco use is a widespread form of addiction affecting the health and lives of millions worldwide annually. This thesis proposes an innovative system based on deep learning techniques for automatic and objective detection of tobacco use. The proposed system consists of two main stages. In the first stage, a robust deep learning model using MediaPipe Pose is developed to detect hand-mouth interactions from video or image data. This model can accurately detect these interactions in real-time. In the second stage, a separate deep learning model is designed to estimate and classify hand movements. This model identifies hand movements by detecting key points of the hand skeleton. The hand movement estimation model can classify specific hand movements associated with smoking behavior (e.g., holding a cigarette, bringing it to the mouth, inhaling) with high accuracy. The performance of the developed system has been evaluated through comprehensive experiments and tests. Tests conducted on different dataset demonstrate that the proposed approach can detect tobacco use with high accuracy rates (above 95%). Moreover, the system's ability to operate in real-time and provide fast response times offers a significant advantage for practical applications. The thesis presents the technical details of the proposed system, including the deep learning architectures used, datasets, preprocessing steps, data augmentation techniques, and experimental results. Additionally, potential future applications of the system, its impact on smoking cessation efforts, and possible improvements are discussed.