摘要
针对传统人体动作识别方法存在的硬件成本高、系统搭建复杂等问题,提出了一个利用普通的商用WiFi设备的人体动作识别系统,通过分析WiFi信号的信道状态信息(CSI)识别家居环境中8个常见的人体动作.为了获取相同时刻的CSI测量值,提出针对不同时间间隔的CSI序列进行插值处理的方法.通过分析不同的子载波和人体动作的相关性,提取不同动作对应的子载波特征方差,进而采用基于稀疏表示分类的算法进行分类.在真实的家居环境中对该系统进行实验,平均识别率可达到96.4%.
Aiming at the problems of high hardware cost and complex system construction in traditional human body activity recognition methods. The human activity recognition system using commercial wireless fidelity(WiFi) devices was presented,utilizing the WiFi signal channel state information(CSI) to identify eight familiar activities. To obtain CSI measurements at the same time,a new method of interpolating CSI sequences for different time intervals was proposed. By analyzing different subcarrier and the correlation of human motion,the variance of subcarrier characteristic corresponding to different actions is extracted,and to recognize it when using the algorithm based on sparse representation classification. Experiment with the system in a real home environment shows the average recognition rate can reach 96. 4%.
作者
肖玲
潘浩
XIAO Ling;PAN Hao(College of Computer Science and Electronic Engineering,Hunan University,Changsha 410082,China;Laboratory of Embedded Systems and Networks,Hunan University,Changsha 410082,China)
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
2018年第3期119-124,共6页
Journal of Beijing University of Posts and Telecommunications
基金
国家自然科学青年基金项目(61300219)
关键词
WIFI
信道状态信息
稀疏表示分类
人体动作识别
wireless fidelity
channel state information
sparse representation classification
human activity recognition