摘要
为进一步提升人体步态识别的准确率,参考人体步态特点,选择下肢表面肌电信号(SEMG)、髋关节角度、膝关节角度作为步态识别信息源,提出一种基于多核相关向量机(MKRVM)的人体步态识别方法.该方法以多源信息特征值作为多核相关向量机的输入,通过实验对不同信号选取合适的核函数,利用萤火虫优化(GSO)算法确定核函数参数,输出为不同步态的概率.利用训练好的模型直接对新样本进行分类,将概率最高的步态模式作为识别结果.实验结果表明,该方法对于平地行走、上楼、下楼、上坡、下坡等步态的平均识别率为94.64%,优于单核支持向量机(SVM)等方法.
A Locomotion-Mode recognition method based on multiple kernel relevance vector machine（MKRVM）was proposed to improve recognition accuracy,which selected the surface electromyography（SEMG）,hip joint angle and knee joint angle as the major information source of recognition according to the user＇s locomotion modes characteristics.SEMG features and joint angle features were fused into a feature vector as the input of multiple kernel relevance vector machine learning model,and different kernel functions were chosen for each signal through experiment.Glowworm swarm optimization algorithm was used to optimize kernel function parameters.The output was the probability of each locomotion mode for this sample.New sample can be classified using the trained model,and the recognition result is the mode with the highest probability.Experiment results show that the average recognition accuracy of locomotionmodes,including level-ground walking,stairs ascent,stairs descent,upslope and downgrade,is 94.64%,which is superior to SVM method using single kernel function.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2017年第3期562-571,共10页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(61174009
61203323)
天津市自然科学基金项目资助项目(13JCQNJC03400)
2016年度河南省高等学校重点科研项目(16B413006)
