基于多特征融合的手势识别研究  

A Hand Gesture Recognition Research Based on Multi-feature Fusion

作　　者：卢梦圆 官巍 马力 LU Mengyuan;GUAN Wei;MA Li(School of Computer Science,Xi'an University of Posts&Telecommunications,Xi'an 710061)

机构地区：[1]西安邮电大学计算机学院,西安710061

出　　处：《计算机与数字工程》2020年第9期2157-2161,共5页Computer & Digital Engineering

摘　　要：基于视觉的手势识别通常采取单一特征用于手势分类,但是单一特征不能代表整个图像。因此论文提出多种特征融合的方法,分别提取了梯度方向直方图(HOG)和局部二值模式(LBP)两种特征,并进行特征融合,融合特征不仅包含图像的局部区域梯度方向信息,还有纹理信息,可以更加全面地描述手势特征。然后将融合特征向量输入SVM分类器完成手势识别。实验表明多特征融合方法相比于单一特征,有更高的识别率。该实验识别了24种手势,最大识别率达98%。Hand gesture recognition methods commonly use a single feature of hand gesture for classification.There is a problem that a single feature cannot represent a whole image.To solve this problem,this paper proposes a hand gesture recognition method based on multi-feature fusion.The extracted features are the gradient direction histogram(HOG)features and the local binary pattern(LBP)features,then,the fusion features are formed.The fusion feature contain not only the information of local gradient magnitude and angle,but also the texture information.Then,a support vector machine is used to train these feature vector.Experimental results show that the multi-feature fusion method has a high recognition rate,compared with the single feature.The proposed model identifies 24 gestures,with a maximum recognition rate of 98%.

关 键 词：手势识别 支持向量机 梯度方向直方图 局部二值模式 特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]