User Recognition Based on Human Body Impulse Response: A Feasibility Study

Human recognition technologies for security systems require high reliability and easy accessibility in the advent of the internet of things (IoT). While several biometric approaches have been studied for user recognition, there are demands for more convenient techniques suitable for the IoT devices. Recently, electrical frequency responses of the human body have been unveiled as one of promising biometric signals, but the pilot studies are inconclusive about the characteristics of human body as a transmission medium for electric signals. This paper provides a multi-domain analysis of human body impulse responses (HBIR) measured at the receiver when customized impulse signals are passed through the human body. We analyzed the impulse responses in the time, frequency, and wavelet domains and extracted representative feature vectors using a proposed accumulated difference metric in each domain. The classification performance was tested using the $k$ -nearest neighbors (KNN) algorithm and the support vector machine (SVM) algorithm on 10-day data acquired from five subjects. The average classification accuracies of the simple classifier KNN for the time, frequency, and wavelet features reached 92.99%, 77.01%, and 94.55%, respectively. In addition, the kernel-based SVM slightly improved the accuracies of three features by 0.58%, 2.34%, and 0.42%, respectively. The result shows potential of the proposed approach for user recognition based on HBIR.