Research On Gesture Recognition Method Based On Single Camera

With the computer and network developing rapidly, human-computer interaction has become an important part in people’s daily life. Some traditional way of human-computer interaction, such as a mouse, keyboard, cannot fully meet the needs of users. Some new human-computer interaction has become a hot topic, such as gesture recognition, speech recognition, face recognition and so on. Compared to other interaction, the gesture, as a means of human-computer interaction, is more natural and direct. It has wide application prospect in intelligent system, robot control, multimedia, entertainment etc..This study is the virtual input method based on hand gesture recognition under monocular vision, involved in the field to the hand gesture segmentation, fingertip detection, hand tracking and gesture recognition. The research mainly includes the following aspects:1. In the part of gesture segmentation, it studied the present common color space model, analyzed the advantages and disadvantages of YCbCr color space. Combined of skin color and motion information, it proposed an adaptive Gauss modeling method based on adjacent frame for skin color detection, and got better effect of segmentation.2. In the fingertip detection, it studied the fingertip detection method based on curvature, screened the fingertips by K-cosine algorithm and K-medoids clustering algorithm. To improve the accuracy of the selected candidate points, it presented a method to adaptive computation of K on K-cosine algorithm. Through the experiment, it improved the detection accuracy.3. In the hand gesture tracking stage, it used the classic Camshift algorithm. Because it was semi-automatic, we need to initialize the target window. In this dissertation, we made the open hands as the starting action, and passed the palm area to the CamShift algorithm as the initial window.4. In the step of gesture recognition, this dissertation selected several obvious characteristic values to form the characteristic vector, and using template matching method for classification and recognition. For the dynamic hand gesture, it used a simple identification, DTW algorithm.Finally, this dissertation made an interactive system for gesture recognition, using dynamic hand gestures to control picture browsing and operation in Picasa, good results were obtained.