基于人体多姿态识别的被动入侵检测模型研究  

Research on passive invasion detection model based on multi-state recognition of human body

作　　者：谷敏敏 刘进军[1,2] 安宁[1] 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009 [2]滁州学院计算机与信息工程学院,安徽滁州239000

出　　处：《传感器与微系统》2015年第6期17-20,共4页Transducer and Microsystem Technologies

基　　金：安徽省高校自然科学基金资助项目(KJ2013B182);教育部高等学校学科创新引智计划资助项目(B14025)

摘　　要：被动入侵检测技术在火灾搜救、空巢老人监护、矿井安全等领域有着广泛的应用。传统的基于无线传感器的被动入侵检测技术大多采用单层链路的信号变化来识别有人入侵和无人入侵两种状态,检测状态过于单一,对于一些人体高危状态如跌倒状态并不能有效的识别。为解决这一问题,提出了一种基于人体多姿态识别的被动入侵检测模型,该模型建立多层信号链路进行特征对比,并使用基于滑动窗口的数据流特征提取方法,实现对各人体入侵姿态的检测。采用多种多分类器算法进行仿真分析,并进一步对人体多姿态的定位和追踪展开实验。结果表明:该检测模型对于正常(无人入侵)、站立、跌倒静躺这三种状态具有很好的识别效果。Passive invasion detection technology has been widely used in fire rescue,empty nesters care,mine safety and other areas. The traditional invasion detection technology based on wireless sensors identification technology mostly only identify whether someone invades,but some high-risk states of human body such as falling can not effectively identified. To solve this problem,propose a passive invasion detection model based on multi-state recognition of human body,build multi-layer signal links to compare each features of data set and use data flow feature extraction method based on sliding window,this model achieves invasion states detection of various human body. Adopt a variety of multi-classification algorithm to simulate and analyze,and further carry out experiment of human body multi-state positioning and tracking. The results show that the detection model has good recognition effect on normal（ nobody）,standing,falling down and lying down states.

关 键 词：射频 多姿态 入侵检测 多分类 

分 类 号：TP212.6[自动化与计算机技术—检测技术与自动化装置]