Hand Detection And Hand Gesture Recognition Research Based On Sequential Images

In the traditional human-computer interaction, the interaction between computers and users is difficult. With the development of computer vision technology, some novel methods have been proposed and make HCI(human-computer interaction) easier than before, for instance, gesture as an interaction method is one of them. However, the hand gesture recognition confronts many challenging problems, such as same hand gesture can be represented by different styles, the interference of complicated background.In the paper, we analyse the problems of the hand gesture recognition, and develop a new recognition system. Our experiments demonstrate the improved system effectively works on the public database. The main work of this paper including:1. For segmenting the hand region from the background, we combine the ViBe algorithm and skin detection algorithm, and obtain the movement information and skin color information, respectively. The hand candidate regions are obtained from this information, and analysed by their connected areas. Finally, a number of hand regions are chosen. Experiments show that this method can detect the hand area accurately, and meet the requirement of real-time.2. In this paper, a hand tracking framework is built. Firstly, Using the Haar-like features and AdaBoost algorithm to select a hand region from several candidate regions;secondly, we combine with CamShift and Kalman filter to track this hand region, where the Kalman filter is used for overcoming the interference from the background.Experiments show that the method can achieve a better accuracy of hand tracking.3. After obtaining the hand gesture, the HOG algorithm is used to obtain the feature of hand gesture, and Random Forest is employed as classifier. HOG can effectively reduce the environment noise. Subsequently, we build a hand gesture system based on the common network camera, combining with the hand detection method and the hand tracking method. Experiments show that the system can identify the hand gesture of different persons, and meet the requirement of real-time, and also achieved a better accuracy of hand recognition.