Abstract:
Usage of rearm by only original users is one of the prime concerns of the research community considering limitless damage and even lethal consequences in case of having the usage in any other way. However, a low-cost, limited-resource, and high-accuracy solution for performing real-time user identi cation of rearm is yet to be proposed in the literature. As a remedy to this situation, in this study, we propose a novel solution that can identify users of a rearm in real time using only a small number of low-cost and low-power COTS (commercial o -the-shelf) pressure sensors. Here, we propose judicious positioning of the sensors such that the number of required sensors can retain a small value ( ve in our case). Besides, we develop a novel machine learning technique, namely Bounded K-means clustering, which exhibits high accuracy in user authentication demanding a small amount of resource and execution time. We evaluate e ciency of the machine learning technique the approach using real data collected from fty users. Our rigorous analysis over the data con rms e ective identi cation of users of a rearm.
Later, using our devised machine learning technique, we develop and implement an operational rearm for identifying its users in real time. The operational rearm consists of only ve COTS pressure sensors at the designated locations as identi ed in the earlier analysis. We present the operational rearm to a new set of users and let them use it in both laboratory setup and day-to-day life. We analyze performance of the rearm over the usage and con rm signi cant accuracy in identifying the users. Further, we generate large-scale synthetic data and perform analysis in the data to con rm scalability of our proposed system. Afterwards, we present an empirical modeling so that any new operational rearm can be tuned for implementing our proposed system. We perform a nal user evaluation based on the ndings of large-scale data analysis and empirical modeling. This evaluation demonstrates e cacy of our system and modeling in parallel to reconciling our results obtained from both earlier user evaluations and synthetic data analysis
