摘要
研究了一种基于多级分类模型的非特定人走路模式识别算法,实现了对水平行走和上、下楼梯三种运动状态的识别.将装有微型加速度传感器的无线数据采集装置固定于人体后腰部,获取运动时的三维步态加速度信号.采用离散小波变换提取与运动相关频带的时频特征,并结合步频以及垂直方向和前进方向加速度信号之间的互相关性,经过特征融合设计了多级分类识别算法.通过对10个人共360组数据的测试结果表明:在步频范围扩大到1～3Hz时,识别率达到了96.1%,且对测试对象的依赖性小.
This paper presented the multiple classifier based walking pattem recognition algorithm, which could identify three walking patterns： horizontal walking, up and down staircase walking. Three-dimensional accelerations during walking were acquired from the wireless accelerometer device fixed on the back waist. The discrete wavelet transformation was applied for time-frequency analysis. The time-frequency features associated with the main frequency band of the motion, walking cadence and the correlation between the vertical and forward acceleration signals were combined to design a multiple classifier. A set of 360 gait samples involving 10 people were used for test,giving an overall recognition accuracy for 96.1% when the walking cadence range was within 1 - 3Hz,and this algorithm was less dependent on individuals.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2009年第8期1794-1798,共5页
Acta Electronica Sinica
基金
国家自然科学基金(No.60673095)
教育部科学技术重点项目(No.208021)
关键词
走路模式
小波变换
时频特征
多级分类器
加速度传感器
walking pattern
wavelet transformation
time-frequency feature
multiple classifier
accelerometer
