Multiple Person Detection by Footsteps Sounds Using GMRS

This article targets footsteps sounds produced by multiple persons aiming to detect the multiple persons separately contained in the sound. To the best of authors' knowledge, there have been no research on the detection of multiple persons from the footsteps sounds. For data processing, we propose to apply the geometric mean for real-time serial-to-parallel converter (GMRS). GMRS is a period estimation method applicable to a composite signal consisting of multiple periodic signals, such like a footsteps sound produced by multiple persons. In particular, GMRS is capable of estimating the period of each of the periodic signals while large power differences exist among the periodic signals. This article provides the performance evaluation of GMRS through computer simulations for the detection of multiple persons from the footsteps sounds as well as experimental results. As a result, it is confirmed that GMRS successfully detects 3 persons.