An experimental study on footsteps sound recognition as biometric under noisy conditions

Abstract

The experimentation of footsteps as a biometric has a short history of about two decades. The process of identification of a person is based on the study of footstep signals captured when walking over a sensing area, and the registering of sounds, pressure, vibration, or a combination of these measures. Application of this biometric can emerge in security systems, that identify persons who enter or leave a space, and in providing help to elderly and disabled persons. In this paper, we are focused in the exploration of pure audio signals of footsteps and the robustness of a person’s classification under noisy conditions. We present a comparison between four wellknown classifiers and three kinds of noise, applied at different signal to noise ratio. Results are reported in terms of accuracy in the detection an users, showing different levels of sensibility according to the kind and level of noise.