基于可穿戴惯性传感技术的人体步态阶段识别  

Human gait phase recognition based on wearable inertial sensing technology

作　　者：陈斯琪 寇俊辉 陈小路 吴铭渝 付国荣 郭良杰[1,4] CHEN Siqi;KOU Junhui;CHEN Xiaolu;WU Mingyu;FU Guorong;GUO Liangjie(China University of Geosciences(Wuhan),Wuhan 430074,China;Hubei Provincial Natural Disaster Emergency Technology Center,Wuhan 430064,China;Yantai Automobile Engineering Professional College,Yantai 265500,China;Engineering Research Center of Rock-Soil Drilling&Excavation and Protection,Ministry of Education,Wuhan 430074,China)

机构地区：[1]中国地质大学(武汉)工程学院,湖北武汉430074 [2]湖北省自然灾害应急技术中心,湖北武汉430064 [3]烟台汽车工程职业学院,山东烟台265500 [4]岩土钻掘与防护教育部工程研究中心,湖北武汉430074

出　　处：《安全与环境工程》2024年第4期11-19,36,共10页Safety and Environmental Engineering

基　　金：湖北省安全生产专项资金科技项目(SJZX20230904);武汉市科技局知识创新专项曙光计划项目(2022020801020209);中央高校基本科研业务费专项资金项目。

摘　　要：为了实现基于可穿戴惯性传感技术的人体步态阶段识别,开发了基于特征选择的人体步态阶段识别模型、基于时间比例优化的人体步态阶段识别模型和基于机器学习多数据类型、多特征、多分类器的人体步态阶段识别模型,并对比了3种模型的步态阶段识别效果。结果表明:基于特征选择的人体步态阶段识别模型的平均识别准确率为73.66%;基于时间比例优化的人体步态阶段识别模型的平均识别准确率为90.96%;利用脚背处俯仰角数据和加速度数据训练得到的基于机器学习的人体步态阶段识别模型的平均识别准确率分别为97.04%、86.80%;针对不同的步态阶段和使用场景,可差异化选择不同的识别方法以获得理想的识别效果;综合采用时间比例优化算法和机器学习方法可以获得较高的综合识别准确率。该研究可为进一步开展基于可穿戴式传感器的人体行为相关研究提供参考。In order to realize recognition of human gait phases based on wearable inertial sensing technology,the human gait phase recognition models based on feature selection,time proportion optimization,and machine learning with multiple data types,multiple features,and multiple classifiers were developed to recognize the human gait phases,and the recognition effect of the three models are compared.The results show that the average accuracy of human gait phase recognition based on feature selection is 73.66%,on time proportion optimization is 90.96%,and on machine learning models trained with pedal pitch angle data and acceleration data is 97.04%and 86.80%,respectively.Different recognition methods can be selectively used according to different human gait phases and application scenarios to achieve desired recognition effects.The comprehensive use of time proportion optimization algorithm and machine learning methods can achieve high comprehensive recognition accuracy.The paper provides a reference for further research on human behavior based on wearable sensors.

关 键 词：人体步态阶段识别 可穿戴惯性传感技术 特征选择 时间比例优化 机器学习 

分 类 号：X912.9[环境科学与工程—安全科学]