A simple vehicle classification framework for wireless audio-sensor networks

Vehicle tracking is one of the important applications of wireless sensor networks. We consider an aspect of tracking: the classification of targets based on the acoustic signals produced by vehicles. In this paper, we present a naive classifier and simple distributed schemes for vehicle classification based on the features extracted from the acoustic signals. We demonstrate a novel way of using Aura matrices to create a new feature derived from the power spectral density (PSD) of a signal, which performs at par with other existing features. To benefit from the distributed environment of the sensor networks we also propose efficient dynamic acoustic features that are low on dimension, yet effective for classification. An experimental study has been conducted using real acoustic signals of different vehicles in an urban setting. Our proposed schemes using a na¨ive classifier achieved highly accurate results in classifying different vehicles into two classes. Communication and computational costs were also computed to capture their trade-off with the classification quality.