摘要
传统手势识别通过提取手势轮廓或关节点位置来进行手势分类识别,这些特征通常在多角度因素下难以准确地表征正确的手势信息,从而导致识别率下降。为了解决目前手势识别在多角度因素导致识别准确率下降的问题,该文提出一种融合HOG(方向梯度直方图)特征和手部多特征的手势识别方法。在特征提取中,首先对处理后的手势图像提取HOG特征并使用PCA降维,然后对手势轮廓提取手部多特征,再将两种特征进行归一化处理后串联融合形成最终的分类特征,最后将最终分类特征通过SVM(支持向量机)进行分类识别,在多角度因素下,该方法能够更准确实时地实现手势识别,平均识别率达到96%,有效地解决了在多角度因素下传统手势识别方法精度不高的问题。
Traditional gesture recognition performs gesture classification by extracting gesture contour or joint position,which is difficult to accurately represent the correct gesture information under multi-angle factors,resulting in the decrease of recognition rate. In order to solve the problem that the recognition accuracy of gesture recognition is reduced due to multi-angle factors at present,a gesture recognition method that fuses HOG(directional gradient histogram)feature and multi-feature of hand is proposed. In feature extraction,the first gesture after the processing image HOG feature extracting and using PCA dimension reduction,then the gesture contour extraction hand more characteristic,then two kinds of characteristics of normalized processing series convergence formed after the final classification characteristics,finally will eventually classification characteristics by the SVM(support vector machine) to identify classification,under the factors from multiple perspectives,the method can achieve more accurate real-time gesture recognition,the average recognition rate of 96%,effectively solved under factors from multiple perspectives,the problems of traditional method of gesture recognition accuracy is not high.
作者
杨述斌
潘伟
蒋宗霖
YANG Shu-bin;PAN Wei;JIANG Zong-lin(School of Electrical and Information Engineering,Wuhan Institute of Technology,Wuhan 430205,China;Hubei Provincial Key Laboratory of Intelligent Robot,Wuhan 430205,China)
出处
《自动化与仪表》
2020年第8期47-51,76,共6页
Automation & Instrumentation
基金
智能机器人湖北省重点实验室开放基金项目(HBIR201406)。
关键词
手势识别
多角度
方向梯度直方图
手部多特征
支持向量机
gesture recognition
multiple perspectives
histogram of oriented gradient(HOG)
hand multifeature
support vector machin
