正在加载图片...
第10卷第3期 智能系统学报 Vol.10 No.3 2015年6月 CAAI Transactions on Intelligent Systems Jun.2015 D0:10.3969/j.issn.1673-4785.201405049 网络出版地址:http://www.cnki.net/kcms/detail/23.1538.tp.20150603.0901.001.html 改进高斯核函数的人体姿态分析与识别 林海波,王浩,张毅 (重庆邮电大学智能系统及机器人研究所,重庆400065) 摘要:为了提高人体动作姿态的识别率,利用Kinect平台构建人体骨骼模型,提出一种基于关节角度的人体姿态 表示方法。同时针对传统的高斯核函数中采用欧氏距离计算方法难以完全反映人体关节运动数据样本点与测试点 之间位置关系的问题,提出了改进的高斯核函数多类支持向量机(MSVM)人体动作姿态识别方法。在高斯径向基核 函数中使用测地线距离代替欧氏距离,建立了基于测地线距离的姿态核函数,采用二叉树方法构建多类支持向量机 完成12种上肢姿态的分类。实验结果表明,该算法取得了较好的识别效果,能更加有效识别人体姿态。 关键词:人体动作姿态;识别;高斯核函数;Kinect:欧氏距离;测地线距离;支持向量机 中图分类号:TP391.9文献标志码:A文章编号:1673-4785(2015)03-0436-06 中文引用格式:林海波,王浩,张毅.改进高斯核函数的人体姿态分析与识别[J].智能系统学报,2015,10(3):436-441. 英文引用格式:LIN Haibo,WANG Hao,ZHANG Yi..Human postures recognition based on the improved Gauss kernel function J].CAAI Transactions on Intelligent Systems,2015,10(3):436-441. Human postures recognition based on the improved Gauss kernel function LIN Haibo,WANG Hao,ZHANG Yi (Research Center of Intelligent System and Robot,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) Abstract:In this paper,a method based on the joint angles of human postures is proposed in order to improve the human posture recognition rate through building a human skeleton model on the Kinect platform.For the traditional method of human postures recognition,Euclidean distance is used in Gaussian kernel function,but the positional relationship of sample point and test point of human body joint can not be reflected completely.So the method of im- proved Gaussian kernel function and multi-class support vector machines (MSVM)is proposed.Using the geodesic distance instead of the Euclidean distance in the Gaussian radial basis kernel function,a posture kernel function based on the geodesic distance is established.Using the binary tree method,a multi-class support vector machine is built to complete classification of 12 kinds of upper limb postures.Experimental results showed that the improved al- gorithm can identify body postures more effectively than before,achieving a good recognition effect. Keywords:human postures;recognition;Gauss kernel function;Kinect;Euclidean distance;geodesic distance; support vector machines (SVM) 基于视觉的人体动作姿态估计和识别在高级人 机交互、智能视频监控、三维动画合成、电影特技制 作、体育运动分析、医学理疗等多个方面拥有十分广 收稿日期:2014-05-22.网络出版日期:2015-06-03. 基金项目:科技部国际合作项目(2010DFA12160):重庆市工业振兴 阔的应用前景,正逐渐受到越来越多学者们的关注。 专项资金资助项目(渝财金[2013]442号). 人体动作姿态分析主要分为2个部分:人体姿态描 通信作者:王浩.E-mail:haoziwang1990@126.com-第 10 卷第 3 期 智 能 系 统 学 报 Vol.10 №.3 2015 年 6 月 CAAI Transactions on Intelligent Systems Jun. 2015 DOI:10.3969 / j.issn.1673⁃4785.201405049 网络出版地址:http: / / www.cnki.net / kcms/ detail / 23.1538.tp.20150603.0901.001.html 改进高斯核函数的人体姿态分析与识别 林海波,王浩,张毅 (重庆邮电大学 智能系统及机器人研究所,重庆 400065) 摘 要:为了提高人体动作姿态的识别率,利用 Kinect 平台构建人体骨骼模型,提出一种基于关节角度的人体姿态 表示方法。 同时针对传统的高斯核函数中采用欧氏距离计算方法难以完全反映人体关节运动数据样本点与测试点 之间位置关系的问题,提出了改进的高斯核函数多类支持向量机(MSVM)人体动作姿态识别方法。 在高斯径向基核 函数中使用测地线距离代替欧氏距离,建立了基于测地线距离的姿态核函数,采用二叉树方法构建多类支持向量机 完成 12 种上肢姿态的分类。 实验结果表明,该算法取得了较好的识别效果,能更加有效识别人体姿态。 关键词:人体动作姿态;识别;高斯核函数;Kinect;欧氏距离;测地线距离;支持向量机 中图分类号:TP391.9 文献标志码:A 文章编号:1673⁃4785(2015)03⁃0436⁃06 中文引用格式:林海波,王浩,张毅. 改进高斯核函数的人体姿态分析与识别[J]. 智能系统学报, 2015, 10(3): 436⁃441. 英文引用格式:LIN Haibo, WANG Hao, ZHANG Yi. Human postures recognition based on the improved Gauss kernel function [J]. CAAI Transactions on Intelligent Systems, 2015, 10(3): 436⁃441. Human postures recognition based on the improved Gauss kernel function LIN Haibo, WANG Hao, ZHANG Yi (Research Center of Intelligent System and Robot, Chongqing University of Posts and Telecommunications, Chongqing 400065, China) Abstract:In this paper, a method based on the joint angles of human postures is proposed in order to improve the human posture recognition rate through building a human skeleton model on the Kinect platform. For the traditional method of human postures recognition, Euclidean distance is used in Gaussian kernel function, but the positional relationship of sample point and test point of human body joint can not be reflected completely. So the method of im⁃ proved Gaussian kernel function and multi⁃class support vector machines (MSVM) is proposed. Using the geodesic distance instead of the Euclidean distance in the Gaussian radial basis kernel function, a posture kernel function based on the geodesic distance is established. Using the binary tree method, a multi⁃class support vector machine is built to complete classification of 12 kinds of upper limb postures. Experimental results showed that the improved al⁃ gorithm can identify body postures more effectively than before, achieving a good recognition effect. Keywords:human postures; recognition; Gauss kernel function; Kinect; Euclidean distance; geodesic distance; support vector machines (SVM) 收稿日期:2014⁃05⁃22. 网络出版日期:2015⁃06⁃03. 基金项目:科技部国际合作项目( 2010DFA12160);重庆市工业振兴 专项资金资助项目(渝财金[2013]442 号). 通信作者:王浩. E⁃mail: haoziwang1990@ 126.com. 基于视觉的人体动作姿态估计和识别在高级人 机交互、智能视频监控、三维动画合成、电影特技制 作、体育运动分析、医学理疗等多个方面拥有十分广 阔的应用前景,正逐渐受到越来越多学者们的关注。 人体动作姿态分析主要分为 2 个部分:人体姿态描
向下翻页>>
©2008-现在 cucdc.com 高等教育资讯网 版权所有