Research On Algorithm For Gesture Recognition Of Smoking Behavior

Smoking in public indoor spaces has been prohibited in many countries because it not only affects the health of the smokers and people around them, but also increases the risk of fire outbreaks. Last year, our government for the first time puts forward studying and making the law that no smoking in public places and fully implementing it within three years. However, because of the human limits, it is difficult to achieve all-round monitoring at any time if there is indoor smoking behavior. This thesis proposes a visual analysis scheme to detect and recognize smoking events by using the gesture of smokers, compared with traditional methods of personnel monitoring, this text method is more real-time and efficiency.The key of research is the gesture area image segmentation and feature extraction about Smoking behavior; this is also the basis of gesture recognition. The paper separately optimize the algorithm from the video object segmentation and feature extraction, can improve operation efficiency, enhance the applicability, finish the gesture recognition in smoking video better.To the gesture area image segmentation, the biggest difference between the human hand of smoking behavior and other activities is that the hand contacts the face at uncertainty period, besides the location of the hand gesture and the overall shape of it. Therefore, the paper presents object segmentation based on combine motion information with skin color detection twice in smoking events, extract the accurate gesture area, and set up a smoking gesture image sample library.Through the analysis and comparison of commonly used method of gesture recognition, according to characteristics of smoking this specific behavior. In recognition phase, the thesis proposes a scheme to detect and recognize the gesture of smoking events by combining Hu moment invariants and edge orientation histogram. Using support vector machine classifier (Lib-SVM)to achieve classification and recognition, the gesture feature vectors from smoking and drinking, eating, scratching, calling, and this similar smoking are collected for training and testing. In order to analyze the effectiveness of the proposed method, the video are captured under indoor environment. The experimental results prove the feasibility of our proposed method...