融合关联性的多任务压缩感知行为识别方法.pdfVIP

CN 43-1258/ TP 第 37 卷第 6 期 2015 年 6 月 计算机工程与科学 ISSN 1007- 130X Computer Engineering . Science Vo l. 37 . No. 6 .]un. 2015 文章编号: 1007- 130X(2015)06-1071-08 融合关联性的多任务压缩感知行为识别方法. 段梦琴，李仁发， 黄 晶 〈湖南大学嵌入式与网络计算湖南省重点实验室，湖南妖沙 410082) 摘 要: 基于传感霉的人体行为识别是一个新兴研究领竣，，作为物联网的一项重要应用，在医疗监护、 助老助残、智能办公/家}[;等方面有着广阔的应用前景. 识别牟是行为识别的一个重要衡量指标，而特征 和分类算法又是影响识别率的两个重要因素.提取了基于多传感器行为识别架构的关联特征，并引入压 缩感知和稀疏表示理论，提出一种多任务压缩感知行为识别方法。最后， 在基准数据库土采用个体无关的 留一验证方法进行了大量实验，结果表明所提出的融合关联性的多任务压缩感知行为识~I) 方法能有效提 升行为识别率，与对应的单任务行为识别方法相比，识J) 速度提高约 56% • 关锺词: 机器学习;物联网;体模网;行为识别;特征提取 中固分类号:TP391 文献标志码:A doi: 10. 3969/j. issn. 1007- 130X. 2015.06.007 Fusing correlation based multi-task compressive sensing for activity recognition DUAN Meng-qin ,Ll Ren-fa ,H UANG Jing (Laboratory of Embedded Systems . Net working. Hunan University. Changsha 410082. China) Abstract: Sensor-based human activity recognition is an emerging research field . It is an important application of Int ernet of Things (loT) , and has very promising application prospects in health care/陀 covery , elder/invalid people assistant , smart home/oHice , etc. Accuracy is one of the most important e- valuation standards of activity recognitíon , and appropriate feature s and classifíers are important accura- cy factors. W e first extract a novel feature called correlation feature. By combining the theory of com- pressive sensing a nd sparse representation , we propose a multi-task compressíve sensing method and use it as the classifier to resolve the problem of actívity recognítion. Finally , we conduct a la