Event classification for living environment surveillance using audio sensor networks

The audio surveillance is traditionally performed by using wired microphones. We propose an alternative surveillance system by using audio sensor networks for event classification in our living environment. We first compare two classical acoustic features - the Fast Fourier Transform (FFT) based acoustic features and the Mel-Frequency Cepstral Coefficient (MFCC) based acoustic features, and then, by using the FFT based acoustic features, we present a hierarchical classification approach for distinguishing abnormal or catastrophic events. A distance based decision fusion is used to combine the sensory information collected by the audio sensors. We present the performance analysis on the proposed approaches using a real audio sensor network.