摘要
文中研究了基于Kinect的手势识别技术,设计并实现了一个功能完善、性能优越的小型手势交互系统。首先结合Kinect获取的人体骨骼信息和深度信息,实现了手部的追踪和提取,并且实验效果不受实验背景、光线、实验者的肤色和服装的影响。然后根据初步获取的手型二值图噪声分布特点,提出一种过滤小规模连通分量像素点的方法对二值图进行去噪。最后,分别以手型二值图Hu矩和手型轮廓二值图Hu矩为特征,使用SVM分类器进行训练和识别。实验结果表明,同手型二值图的Hu矩相比,以手型轮廓二值图的Hu矩作为特征具有明显优势。
It explores the technology of gesture recognition based on Kinect in this paper, and designs and implements a small gesture in- teractive system with perfect function and excellent performance. At first, the Skeleton Information and Depth Information from Kinect is used to track and extract hand from the background. The result doesn＇ t be affected by the background,light, and experimenter＇ s skin col- or and costume. Then, according to the distribution of noise from preliminary acquisition in the binary images, a method of filtering the small scale connected component pixels is put forward to denoise. At last, Hu＇ s moments of hand binary images and hand contour binary images are used as features to train the Support Vector Machine （SVM） classifiers respectively. The experimental results show that com- pared with Hu＇ s moments of hand binary images, the Hu＇ s moments of the hand contour binary images have obvious advantages.
出处
《计算机技术与发展》
2016年第8期200-204,共5页
Computer Technology and Development
基金
国家"863"高技术发展计划项目(2012aa01a509
2012aa01a510)
关键词
静态手势识别
手势交互系统
HU矩
支持向量机
static gesture recognition
gesture interactive system
Hu＇ s moment
support vector machine
