基于改进的YCbCr空间及多特征融合的手势识别  

GESTURE RECOGNITION BASED ON IMPROVED YCBCR SPACE AND MULTI-FEATURE INTEGRATION

作　　者：薛俊韬[1] 纵蕴瑞 杨正瓴[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《计算机应用与软件》2016年第1期151-155,共5页Computer Applications and Software

基　　金：天津市科技支撑计划重点项目(10ZCKF SF01100);天津市科技型中小企业创新基金项目(13ZXCXGX40400)

摘　　要：针对基于视觉的手势识别的复杂性,提出一种基于改进的YCbCr空间及多特征融合的手势识别新方法。首先针对YCbCr颜色空间易受环境因素影响的特点,采用改进的YCbCr椭圆聚类肤色模型的手势分割方法提取手势区域;然后按手势图像外接矩形的宽高比和手指个数进行粗分类,再提取手势的Hu矩和傅里叶描述子构建融合特征,并将融合特征输入BP神经网络进行训练识别;最后综合粗分类和BP神经网络的结果进行手势判别。实验结果表明,该方法在保证实时性的同时具有较高的识别率。Because of the complexity of vision-based hand gesture recognition, we presented a novel hand gesture recognition algorithm which is based on improved YCbCr space and multi-feature integration. Firstly, considering the characteristic that YCbCr colour space is prone to the influence of environmental factors, the algorithm adopts the improved hand gesture segmentation method using YCbCr elliptic clustering skin colour model to extract hand gesture region. Then it makes initial classification according to the aspect ratio of envelop rectangle of hand gesture image and the number of fingers, and extracts Hu moment and Fourier descriptor of hand gesture to build integration features, which are put into BP neural network for training and recognition. Finally the results of the initial classification and BP neural network are combined for hand gesture recognition. Experimental results showed that the proposed method could ensure the real-time performances while getting a quite higher recognition rate.

关 键 词：手势识别 YCBCR 颜色空间 HU矩 傅里叶描述子 BP神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]