Acoustic and seismic signal processing for footsetp detection

Abstract

The problem of detecting footsteps using acoustic and seismic sensors is approached from three different angles in this thesis. First, accelerometer data processing systems are designed to make footsteps more apparent to a human operator listening to accelerometer recordings. These systems work by modulating footstep signal energy into the ear's most sensitive frequency bands. Second, linear predictive modeling is shown to be an effective means to detect footsteps in accelerometer and microphone data. The time evolution of the third order linear prediction coefficients leads to the classical binary hypothesis testing framework. Lastly, a new method for blindly estimating the filters of a SIMO channel is presented. This method is attractive because it allows for a more tractable performance analysis.