Indexed by:Conference Paper

Date of Publication:2016-10-04

Included Journals:Scopus、SCIE、CPCI-S、EI

Page Number:418-422

Key Words:Feature selection; Online learning; Activity recognition; Big data analytics

Abstract:Recently, sensor based human activity recognition has attracted lots of researches, which overcomes the constraint of the traditional vision-based technology. Fall detection is an important part of monitoring elderly people. In order to fast and precisely detect the action fall from huge data, online feature selection method is proposed to improve online human activity recognition, which overcomes the limit of traditional batch learning methods. The high-dimensional real-world dataset is utilized to evaluate the proposed method. Better results have demonstrated the efficacy of the proposed online feature selection approach. It has also been demonstrated that data from belt is much more easily discriminated than from other three body parts.