摘要
在现有支持向量机多分类方法基础上,提出了一种改进的有向无环图支持向量机(DAGSVM)手势识别方法.分析了传统有向无环图支持向量机分类器生成顺序随机化的不足,引入类间距离和类的标准差作为生成分类器的测度.利用Kinect获取场景深度信息得到手势图像,提取手势特征并训练SVM分类器,并采用改进后的方法得到DAGSVM分类器.实验证明:与其他支持向量机多分类器相比,改进后的DAGSVM分类器能够达到更高的识别率,将这个手势识别方法用于智能轮椅的控制上,取得了良好的效果.
On the basis of SVM(support vector machine) multiclass classification,an improved DAGSVM(directed acyclic graph support vector machi ne) hand gesture recognition approach was put forward.The traditional DAGSVM′s randomized generating sequence was insufficient.For this reason,the distance among two classes and standard deviation were introduced.The depth information of the scene by kinect was obtained and feature vectors used to train multiple b inary SVM classifiers were extracted.Then DAGSVM was constructed by the method introduced.Finally,the experimental results prove that the improved DAGSVM can reach higher recognition rate and used in the control of intelligent wheelchair s successfully.
出处
《华中科技大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2013年第5期86-89,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
科技部国际合作资助项目(2010DFA12160)
国家自然科学基金资助项目(51075420,60905066)
重庆市科技攻关项目(2010AA2055)
关键词
智能轮椅
手势识别
人机交互
有向无环图
支持向量机
深度信息
intelligent wheelchair
hand gesture recogniti on
human computer interaction
directed acyclic graph(DAG)
support vector mac hine(SVM)
depth information