摘要
为提高多模式人体下肢动作识别的准确性,提出了一种基于简约支持向量机算法的下肢动作识别方法。通过动作分解将人体日常下肢动作行为分解为不同的动作片段以组成识别目标集;以下肢肌肉表面肌电信号为信息源,综合短时统计时域特征值和Mallat小波时频域特征值建立识别特征向量空间;采用核聚类简化的方法降低计算复杂度,提高算法的鲁棒性。起立、平地常速行走以及上下楼梯等四个日常下肢动作识别实验的结果证明了该方法的有效性。
In order to improve precision of the multi-motion pattern recognition of lower limb movements,a recognition method based on reduced SVM was proposed.The daily movements were decomposed into different segments to form the recognition goals,and the surface electromyography signals of low limb muscles were used as information sources.The eigenvectors were extracted using Mallat wavelet transform and statistical characteristics in time-domain.Then the simplified method of kernel clustering was used to reduce the computational complexity and improve the robustness of algorithm.Four patterns were identified herein,i.e.standing,normal speed walking,up-stairs and down-stairs.The experimental results show the effectiveness of this method.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2011年第4期433-438,共6页
China Mechanical Engineering
基金
国家自然科学基金资助项目(60703041)
广东省教育部产学研结合项目(2009B090600139)
浙江理工大学科研启动基金资助项目(1009831-Y)
关键词
下肢动作识别
表面肌电信号
支持向量机
多元分类
lower-limb movement recognition
surface electromyo graphy（EMG）
support vector machine（SVM）
multi-classification
