基于腕部IMU数据采集的人体运动模式识别  

Human Motion Pattern Recognition Based on Wrist IMU Data Collection

作　　者：胡春生[1] 赵汇东 Hu Chunsheng;Zhao Huidong(School of Mechanical Engineering,Ningxia University,Yinchuan 750021,China;The Affiliated High School of Shanxi University,Taiyuan 030006,China)

机构地区：[1]宁夏大学机械工程学院,宁夏银川750021 [2]山西大学附属中学,山西太原030006

出　　处：《宁夏大学学报（自然科学版）》2021年第1期45-50,57,共7页Journal of Ningxia University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61502257)。

摘　　要：利用安卓智能可穿戴设备内置的运动传感器采集了特征频率变化较高的人体腕部加速度信息,对人体行走、跑步、站立、下楼4个运动模式进行了分析,提取了运动传感器信号均值、峰值、方差和协方差等特征值,并利用支持向量机算法对4种运动模式分类识别,准确率为88.46%.随后利用网格搜索法、遗传算法、粒子群优化算法对参数C,g进行寻优,提高准确率至90%以上.结果表明,本方法可有效识别实验者静止站立、走路、跑步、下楼4种动作,为人体运动模式识别提供参考依据,并为进一步识别运动模式提供研究基础.Human wrist acceleration information with high variation of characteristic frequency is collected by using the android smart wearable device with a built-in motion sensor,then 4 movement patterns of human body(walking,running,standing up,going downstairs)are analyzed to extract the motion sensor signal of meanvalue,peak value,variance and covariance characteristics.Besides,four kinds of motion pattern are classified and recognized by using support vector machine(SVM)algorithm,the recognition accuracy is 88.46%.Subsequently,grid search method,genetic algorithms and particle swarm optimization algorithm are used to optimize parameters C and g,this can improve the accuracy to over 90%.The results show that this method can effectively identify 4 movement patterns of walking,running,standing up and going downstairs,providing a reference for the recognition of human movement pattern,and giving a research reference for the further recognition of the movement pattern.

关 键 词：人体运动识别 加速度传感器 陀螺仪传感器 支持向量机 参数优化 智能可穿戴设备 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]