Human Motion Patterns Recognition based on RSS and Support Vector Machines

Sameer Ahmad Bhat, Abolfazl Mehbodniya, Ahmed Elsayed Alwakeel, Julian L. Webber, Khalid Al-Begain

Abstract: In this paper, we propose a novel received signal strength (RSS) and machine learning (ML) based system for recognizing human motion walking patterns. Our proposed system eliminates the need to attach sensors on a human body, thereby enabling a non-invasive monitoring system. The proposed system model exploits the potential of a single channel RSS in precisely identifying unique human motion patterns in indoor environments and implements a support vector machine (SVM) algorithm for higher pattern detection accuracy. We validate our proposed model by developing a testbed setup based on state-of-the-art Software Defined Radios (SDRs) and provide a comparative analysis of machine learning models used in the patterns classification process. Our study results reveal that unique walking patterns embedded within RSS and with machine learning classifier, can precisely help in identifying human motion patterns with detection accuracy of approximately 99 percent. The study results impact research scholars actively engaged in developing human motion recognition systems, intrusion detection systems, or healthcare monitoring systems, and in those developing innovative and efficient techniques for monitoring and control systems.

External IDs:dblp:conf/wcnc/BhatMAWA20