Improving cell phone based gait identification with optimal response time using cloudlet infrastructure

Abstract

In this paper, we propose an improved gait identification based on signal collected from mobile sensors (e.g. accelerometer, magnetometer). Based on the observation from previous works, we found that there are restrictions which could negatively affect the efficiency of the system when it is applied in reality. For example the installation error has never been considered well. Additionally, performing identification tasks on mobile devices with limited resource constraints is also a big challenge. In this paper, we propose our own identification method which achieves better accuracy than previous works by taking a deep look at processing steps in gait identification issue. Moreover, the interaction between our identification model and human interaction is improved by minimizing the time delay to perform identification. To do this, the VM-based cloudlet infrastructure is also constructed to perform assigning computation tasks from mobile to nearby powerful PCs that belong to the cloudlet. From initial experiment, the archived accuracy of our identification model was approximately 98.99 % and the response time was reduced by 95.8 %. © 2013 Springer Science+Business Media Dordrecht.