Distributed classification of acoustic targets in wireless audio-sensor networks Baljeet Malhotra, Ioanis Nikolaidis * , Janelle Harms Computing Science Department, University of Alberta, 221 Athabasca Hall, Edmonton, Alberta, Canada T6G 2E8 article info Available online xxxx Keywords: Sensor networks Audio sensing applications Classification algorithms k-Nearest neighbor abstract Target tracking is an important application for wireless sensor networks. One important aspect of tracking is target classification. Classification helps in selecting particular tar- get(s) of interest. In this paper, we address the problem of classification of moving ground vehicles. The basis of classification are the audible signals produced by these vehicles. We present a distributed framework to classify vehicles based on features extracted from acoustic signals of vehicles. The main features used in our study are based on FFT (fast Fou- rier transform) and PSD (power spectral density). We propose three distributed algorithms for classification that are based on the k-nearest neighbor (k-NN) classification method. An experimental study has been conducted using real acoustic signals of different vehicles recorded in the city of Edmonton. We compare our proposed algorithms with a naive dis- tributed implementation of the k-NN algorithm. Performance results reveal that our pro- posed algorithms are energy efficient, and thus suitable for sensor network deployment. Ó 2008 Elsevier B.V. All rights reserved. 1. Introduction Networked sensors are equipped with various sensing devices, as well as memory, processor, radio, and a power supply. However, they are still constrained with limited memory, processing power, and channel capacity. In gen- eral these sensors are operated on batteries; hence they have limited energy reserves. Energy conservation is one of our design goals. In broad terms, the processing and transmission of data–intensive data sources (e.g. high frame rate/high resolution video) is intrinsically more expensive in terms of energy compared to sensing based on low data rate sources (e.g. audio capture). Ground vehicle tracking is one of the applications that have attracted considerable attention because of its mili- tary and civilian application. Tracking can be achieved if vehicles can transmit identification signals periodically. However, it is unlikely that such arrangements can be made in hostile environments such as for the tracking of unfriendly or enemy vehicles. Tracking can also be facili- tated by video sensors, but at a higher energy cost. This sit- uation motivated us to consider acoustic sensors for tracking ground vehicles. In the rest of the report the terms ‘‘target” and ‘‘vehicle” are used interchangeably. Vehicle tracking may have certain objectives that must be supported, including detection, identification and/or clas- sification, and localization. Depending on specific tracking objectives all or a combination of these elements may be required. We consider vehicle tracking based on the dis- tinctive acoustic signatures produced by different vehicles. Signatures of different vehicles are distinctive because of the sound produced by their engine and propulsion mech- anism is unique [10]. The problem of vehicle detection using the acoustic signature has been the topic of study for several years and many solutions have been proposed [2,11,10]. Target classification in wireless sensor networks is a relatively recent topic of study [1,6,7,4]. Sensor net- works provide an extra advantage: redundant sensors can work together in a coordinated fashion to detect and report the presence of a target vehicle. Once the target has been detected the sensors can inform their neighboring sensors for a continuous update (and possibly continuous refine- 1389-1286/$ - see front matter Ó 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.comnet.2008.05.008 * Corresponding author. Tel.: +1 780 4925757; fax: +1 780 4921071. E-mail addresses: baljeet@cs.ualberta.ca (B. Malhotra), yannis@cs. ualberta.ca (I. Nikolaidis), harms@cs.ualberta.ca (J. Harms). Computer Networks xxx (2008) xxx–xxx Contents lists available at ScienceDirect Computer Networks journal homepage: www.elsevier.com/locate/comnet ARTICLE IN PRESS Please cite this article in press as: B. Malhotra et al., Distributed classification of acoustic targets in wireless ..., Comput.- Netw. (2008), doi:10.1016/j.comnet.2008.05.008