体域网中基于特征组合的步态行为识别  

Gait behavior recognition based on feature combination in body area network

作　　者：王凯[1] 孙咏梅[1] 张泓[1] 武杨[1] 纪越峰[1] 

机构地区：[1]信息光子学与光通信国家重点实验室,北京邮电大学,北京100876

出　　处：《中国科学：信息科学》2013年第10期1353-1364,共12页Scientia Sinica(Informationis)

基　　金：国家重点基础研究发展计划(973计划)(批准号:2011CB302702);国家自然科学基金重大项目(批准号:61190114)资助

摘　　要：物联网(internet of things,IOT)拥有无处不在的识别、传感和通信能力,体域网(body area network,BAN)属于物联网中和人体相关的领域,其应用广泛,可以在日常生活中对人们进行监测及提供帮助.行走是许多日常活动的基本环节,因而步态分析能为体域网应用提供重要的生理行为信息.现有的步态分析已取得一定的研究成果,但仍存在一些问题,例如大多数步态特征提取是对加速度信号进行6重以上的变换,使得特征达到了45维以上,最后需要通过降维或优化来简化特征,较为复杂.本文设计一种灵活便捷的数据采集系统,并利用小波变换、傅里叶变换和四分位差提取出加速度信号中比较简单、低维度但能反应运动特征的步态参数,之后通过模式识别算法进行步态行为识别验证.实验结果表明该系统使用方便,特征提取方法简单实用,识别精确度为97%,EER(equal error rate)最小可到0.9%.The Internet of Things（IOT） has ubiquitous capabilities of identification,sensing and communication.Body Area Network（BAN） belongs to IOT field related to the human body. It is widely used in e-health applications, helping people monitor their activities in their daily lives. Walking is the basic part of many daily activities, hence gait analysis can provide important information about the physiological behavior in BAN applications. Many existing researches on gait analysis have a good results, but there are still some problems. For example, most gait feature extraction are carried out by more than 6 conversions for acceleration signal, so that the characteristic dimension reaches 45 or more. Therefore, they need to simplify the features by dimension-reduction or optimization, which makes them very complex. In this paper, we firstly design a flexible and convenient data acquisition system. Then we use Wavelet Transform, Fourier Transform and Interquartile Range to extract some simple, low-dimensional motion characteristics of acceleration signal to reflect the characteristics of the movement. Finally, we make recognition and verification of gait behavior by pattern recognition algorithms.Experimental results show that the system is easy to use and the method of feature extraction is simple and practical. And its recognition accuracy reaches 97% and minimum EER（Equal Error Rate） reaches 0.9%.

关 键 词：物联网 体域网 加速度 特征组合 步态识别 

分 类 号：TP391.44[自动化与计算机技术—计算机应用技术] TN929.5[自动化与计算机技术—计算机科学与技术]