基于结构分析的手势识别

Hand Gesture Recognition Based on Structure Analysis

该文从分割和表示(建模)两方面着手,提出了一种新颖的手势分割和整体及局部手势特征提取算法.用模糊集合来描述视频流中空域和时域上的背景、颜色、运动等信息,通过对它们执行模糊运算,分割出人手;使用结构分析的方法来表示手势,根据人手不同部分在几何尺寸上的变化,从低到高逐次分析图像金字塔中各种分辨率的图像,以获取手势的整体和局部结构特征;将人手划分成手掌和手指几个部分,使用手掌和各手指的中心点的坐标和从手掌中心到所有手指的中心的方向(作为手势方向)来表示一个2D手势.实验结果证明,该文算法具有很好的鲁棒性,对手势分割中间结果的精确性要求不高,因此能适应环境的变化.

Camera-based user interface is one of the main style in multimodal user interface, which includes gesture interaction, pose interaction, gaze interaction~ and so on. Camera-based gesture interaction is a natural, convenient way to interact with computer. Gesture recognition is the basis of gesture interaction. This paper considers both hand segmentation and gesture modeling, and presents a method of gesture segmentation and globe and local gesture features extraction. The background, motion, and color information of the video is described by fuzzy set, and is processed with fuzzy operation, and then the hand is segmented from the environment. Because different parts of the hand have different size, so Image Pyramid is used to analyze them to obtain the g obe and local characteristics of the gesture. Then the hand is simply divided into palm and five fingers. The coordinates of the center of palm and every finger and the orientation from the center of the palm to the center of all the five fingers （as the orientation of the gesture） are used to denote a 2D gesture. This algorithm can be used to recognize 2D gesture. In experiment, the algorithm is applied to real-time gesture segmentation and recognition successfully, and robust enough against environment noise.