基于Kinect骨骼数据的静态三维手势识别

STATIC THREE-DIMENSIONAL GESTURE RECOGNITION BASED ON KINECT SKELETON DATA

手势作为一种自然、直观的交流方式,在人机交互领域得到越来越广泛的应用。研究的手势是指手与臂形成的一种空间三维姿势,现有方法对该种手势识别的准确性不高且实时性不强。在Kinect体感摄像机获取的人体手部关节点三维坐标基础上,提出一种计算手部角度进行静态三维手势识别的新方法。该方法通过计算手部多个位置的夹角来获取手部形态特征,然后与参考的静态手势特征做匹配识别。实验表明,该方法能够判断和识别当前静态手势与参考手势是否匹配,比现有方法具有更好的识别准确性和更强的实时性。

As a natural and intuitive way of communication, gestures are widely applied to the field of human-computer interaction. The gesture in our research refers to a kind of spatial three-dimensional posture formed by hand and arm, but current methods on recognising such gesture are not accurate and are poor in real-time performance as well. On the basis of 3D coordinates of human hand joints captured by Kinect somatosensory camera, we present a novel method which makes static three-dimensional gesture recognition by calculating the hand angles. The method retrieves the morphological features of hands by calculating several angles between multiple positions of hand, and then carries out the matching recognition using the static gesture features as the reference. Experiments show that our method can judge and recog- nise whether the current static gesture matches the reference one or not, it has higher recognition accuracy and better real-time performance than the existing methods.