Study On Gesture Recognition Technique Based On Apparent Model Under Complex Background

Gesture recognition research has broad prospects for practical applications. Vision based gesture recognition can provide a more natural and harmonious way of human-computer interaction, and it is the trend of gesture recognition technology. However, due to sign is diversity of uncertainty with time and space, so the current way of human-computer interaction is still in its experimental stage. So far the theory of gesture recognition is not very mature, and the scope of signs which can be identified is relatively small, so this field is extremely challenging.Vision-based gesture recognition is composed of three parts. They are gesture segmentation, feature extraction and recognition. On the basis of reading a large number of reference materials, this paper accomplishes the following works.At the stage of the gesture segmentation, this paper compares and analyzes several gesture segmentation algorithm which are commonly been used. After comparing the distribution of the skin color in various color space, we propose a segmentation algorithm which is based adaptive luminance in the YCbCr space, and the experiment proves that the algorithm can complete well in the task of gesture segmentation under complex background.At the stage of the feature extraction, this paper compares and analyzes several feature extraction methods which are commonly been used. Then we use Fourier descriptor and moment invariants as the characteristics of a gesture respectively, and analyze the number of the Fourier descriptor which is more suitable.At the stage of the gesture recognition, we discuss static gesture recognition and dynamic gesture recognition in this paper. In the static gesture recognition, we analyze the method of template matching briefly at the first, then detail the method of neural network and use BP neural network to train and recognize 10 gestures with 600 samples. In the dynamic gesture recognition, the paper study on the key technologies of the dynamic gesture recognition preliminarily. A method of combining the tracking identification with the key frame is proposed, which is proved to be effective and satisfying.