注:国家自然科学基金(NO.61771059);高动态导航技术北京市重点实验室开放课题
作者:马语晗,赵辉
单位:北京信息科技大学高动态导航技术北京市重点实验室,北京 100101
中图分类号:TP391.9
文献标识码:A
文章编号:1006-883X(2018)09-0028-06
收稿日期:2018-04-10
摘要:针对人体运动模式识别中最优识别特征难以确定的问题,提出一种基于Relief-F特征加权支持向量机的运动模式识别算法。选取MEMS惯性传感器的加速度时域特征构成特征向量,通过Relief-F算法对特征向量各元素进行权重估计,构造一个最优权重特征向量,增大不同运动模式间特征向量的差异性,采用支持向量机作为分类器,可识别站立、走、跑、跳、跌倒、上楼、下楼7种运动模式。实验表明,所提出的算法能够准确识别多种运动模式,识别精度可达94.1%。
关键词:扩展过滤式选择算法;支持向量机;惯性传感器;时域特征向量;运动模式。
Motion Pattern Recognition Based on Feature Selection Weighted Support Vector Machine
MA Yu-han, ZHAO Hui
Beijing Key Laboratory of High Dynamic Navigation Technology, Beijing Information Science and Technology University, Beijing 100101, China
Abstract: Aiming at the problem that the optimal recognition feature is difficult to determine in human motion pattern recognition, a motion pattern recognition algorithm based on Relief-F feature-weighted Support Vector Machine (SVM) is proposed in this paper. The time domain features of MEMS inertial sensor are selected to constitute the eigenvectors. Relief-F algorithm is used to estimate the weights of the eigenvectors. An optimal weight eigenvector is constructed to increase the difference of eigenvectors between different motion modes. As a classifier, SVM can identify 7 kinds of sport patterns of standing, walking, running, jumping, falling, upstairs and downstairs. The experimental results show that the proposed algorithm can identify a variety of motion patterns accurately with the recognition accuracy of 94.1%.
Key words: Relief-F; Support Vector Machine (SVM); inertial sensor; time domain feature vector; motion pattern
阅读全文
备注:2018年 第24卷 第09期