基于智能手机内置传感器的人体运动状态识别  

Human motion state recognition based on smart phone built-in sensor

作　　者：殷晓玲[1,2] 陈晓江[1] 夏启寿[1,2] 何娟 张鹏艳 陈峰[1] YIN Xiaoling;CHEN Xiaojiang;XIA Qishou;HE Juan;ZHANG Pengyan;CHEN Feng(School of Information Science and Technology,Northwest University,Xi’an 710127,China;College of Mathematics and Computer Science,Chizhou University,Chizhou 247000,China)

机构地区：[1]西北大学信息科学与技术学院,陕西西安710127 [2]池州学院数学与计算机学院,安徽池州247000

出　　处：《通信学报》2019年第3期157-169,共13页Journal on Communications

基　　金：国家自然科学基金资助项目(No.61170218;No.61272461;No.61373177)~~

摘　　要：针对目前智能手机识别人体运动状态种类少、准确率低的问题,提出一种利用加速度传感器和重力传感器分层识别人体运动状态的方案。首先,利用加速度和重力加速度的关系计算出与手机方向无关的惯性坐标系下的线性加速度;其次,根据人体运动频率的变化范围和线性加速度矢量来确定脚步的波峰和波谷位置;最后,提取线性加速度在时域上的特征向量,使用层次支持向量机方法分层识别人体运动状态。实验结果表明,该方法能有效识别人体6种日常运动状态,准确率达到93.37%。To solve problems of low accuracy and fewer types of human motion state recognized by current smart phones,a method to do hierarchical recognition by using acceleration sensors and gravity sensors was proposed.Firstly,linear acceleration in inertial coordinate system and independent of phone direction was calculated by using the relation between acceleration and gravity acceleration.Secondly,according to the span of human motion frequency and linear acceleration vector,positions of peak and trough of footsteps were determined.Finally,feature vector of linear acceleration in time domain was extracted and human motion states were recognized hierarchically by using hierarchical support vector machine(H-SVM).The experiment shows the method can recognize six usual human motion states,while accuracy rate up to 93.37%.

关 键 词：运动状态识别 层次支持向量机 智能手机传感器 时域特征 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]