基于双源传感与LSTM的运动步态相位识别研究  

Research on phase recognition of motion gait based on dual source sensing and LSTM

作　　者：李鑫 徐赵洁 柳金康 梁旭 LI Xin;XU Zhaojie;LIU Jinkang;LIANG Xu(School of Mechanical and Materials Engineering,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学机械与材料工程学院,北京100144

出　　处：《传感器与微系统》2025年第4期13-16,20,共5页Transducer and Microsystem Technologies

基　　金：国家重点研发计划资助项目(2023YFE0202100);国家自然科学基金资助项目(62103007);北方工业大学科研启动基金资助项目(110051360023XN224-32)。

摘　　要：为了实现人体下肢不同运动模式步态相位更精确的识别,提出一种基于双源传感与长短期记忆(LSTM)的辨识方法。首先,对跑步、平地行走、上坡及上楼4种运动模式下运动学与生物学信息采集,并对4种步态进行了相位的划分。其次,采用经验公式对多运动模式进行特征提取,并利用LSTM算法实现对不同运动模式下的相位辨识,得到4种运动模式步态相位辨识平均准确率为95.76%。通过与K最近邻(KNN)、决策树(DT)算法对比,分别高出2.25%与1.66%。同时,与肌电、惯性测量单元(IMU)单源传感器辨识平均准确率对比,分别高出2.82%与2.70%。验证了在不同运动模式步态相位划分的基础上,双源传感与LSTM算法辨识的有效性。In order to realize more accurate recognition of gait phase in different motion modes of human lower limbs,an identification method based on dual source sensing and long short-term memory(LSTM)is proposed.Firstly,the kinematic and biological information of running,flat walking,uphill and upstairs movement modes are collected,and the phase of the four gaits is divided.Secondly,the empirical formula is used to extract the features of the multi-motion modes,and the LSTM algorithm is used to realize the phase identification under different motion modes.The average accuracy of the phase identification of the four motion modes is 95.76%.Which are 2.25%and 1.66%higher,respectively,compared with K nearest neighbour(KNN)and decision tree(DT)algorithms.Meanwhile,compared with the average identification accuracy of electromyography and inertial measurement unit(IMU)single-source sensors,which are 2.82%and 2.70%higher,respectively.On the basis of gait phase division in different motion modes,identification validity of dual-source sensing and LSTM algorithm is verified.

关 键 词：运动模式 步态相位 肌电传感器 惯性测量单元 长短期记忆算法 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]