基于惯性传感器和LSTM神经网络的人体运动识别方法  

Inertial Sensor and LSTM Neural Network based Human Motion Recognition Method

作　　者：佟丽娜[1] 马航航 彭亮[2] TONG Lina;MA Hanghang;PENG Liang(School of Mechanical Electronic&Information Engineering,China University of Mining and Technology-Beijing,Beijing 100083,China;State Key Laboratory of Management and Control for Complex Systems,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]中国矿业大学(北京)机电与信息工程学院,北京100083 [2]中国科学院自动化研究所复杂系统管理与控制国家重点实验室,北京100190

出　　处：《传感技术学报》2020年第11期1536-1543,共8页Chinese Journal of Sensors and Actuators

基　　金：国家重点研发计划项目(2018YFB1307000);国家自然科学基金项目(U1613228);北京市自然科学基金项目(Z170003)。

摘　　要：面向人体惯性运动捕捉系统,提出一种基于长短期记忆网络(LSTM)的人体运动模式识别方法。设计1个包含2层LSTM层的深度学习神经网络对人体三维加速度信息进行自动特征提取并对多类运动模式进行时序建模,从而实现对运动模式的快速实时识别。实验基于WISDM公开数据集,对人体右腿前口袋部位的三维加速度信息进行分析,验证模型识别人体慢跑、步行、坐、站、上楼、下楼运动过程的准确率为97.5%,并通过对比实验验证了该方法的有效性。为基于移动设备及可穿戴设备的人体运动识别研究提供了一个可行的方法。For human inertial motion capture system,a Long Short-Term memory(LSTM)network based on human motion recognition method is proposed.Here we designed an LSTM deep learning neural network with 2 LSTM layers extract human tri-axal acceleration features automatically and build time series model for multi-motion patterns,so as to recognize human motions real timely.Basing on the WISDM open data set,analyzing the tri-axal acceleration from front pocket of right leg,the experiment showed effectiveness by comparing with other methods,and it can recognize human jogging,walk,sit,stand,upstairs and downstairs with an accuracy of 97.5%.This paper proposes a feasible method for human motion recognition based on mobile and wearable devices.

关 键 词：人体运动识别 惯性运动信息 LSTM神经网络 时序模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]