Published in IET Intelligent Transport Systems Received on 15th October 2011 Revised on 24th April 2012 doi: 10.1049/iet-its.2011.0162 ISSN 1751-956X Comparative performance analysis of three classifiers for acoustic signal-based recognition of motorcycles using time- and frequency-domain features B.S. Anami 1 V.B. Pagi 2 S.M. Magi 2 1 KLE Institute of Technology, Hubli, Karnataka, India 2 Basaveshwar Engineering College, Bagalkot, Karnataka, India E-mail: veereshpagi@yahoo.com Abstract: Vehicles of different types generate dissimilar sound patterns even in similar working conditions. In this study, the motorcycles are classified into bikes and scooters based on the sounds produced by them. Simple time-domain features and frequency-domain features are used for classifiers. The performances of artificial neural network, knowledge-based classifier and dynamic time warping are compared and reported. All these classifiers have shown more than 90% classification accuracy when trained with minimum 40% of the samples. 1 Introduction Motorcycles are the obvious travel companions of middle class Indians. Motorcycles account for nearly 77% of the vehicle sales in India, because of their affordability, road conditions and fuel efficiency ‘(Courtesy: India-Reports on 20 February 2010)’. The motorcycle sales are projected to exceed 10 million units by 2012–13. Sound patterns generated by moving vehicles vary depending on their types and hint information necessary to classify the vehicles. Various sound sources include engine, rotating parts and exhaust system. Other factors influencing variation of sound are condition of the vehicle, environmental conditions, road conditions, maintenance of the vehicle and so on. The work aims to classify the motorcycles into bikes and scooters, based on the emitted acoustic signals. Mopeds are not included in this work because they are almost obsolete. The systems such as speech recognition, speaker recognition, music classification and musical instrument recognition employ acoustic signals. For speech recognition, there are efficient classification algorithms and well-established sound alphabet. However, non-speech sound recognition applications such as environmental sound recognition vehicle classification, fault diagnosis need the development of efficient signal classification algorithms. The present study uses the time- domain features, zero-crossing rate (ZCR), short-time energy (STE) and root mean square (RMS) of the sound signal. The frequency-domain features employed in this work are mean and standard deviation of the spectral centroid (CMean and Csd). The extracted features are input to different classifiers. The artificial neural network (ANN), dynamic time-warping (DTW) and knowledge-based classifier (KBC) are used for classification. The present work serves as the preliminary step for further investigation on fault diagnosis and fault source localisation. The traffic census of the vehicles can be carried out by the findings of the work. The automated census points can be set, which will count the number of vehicles, in general, that pass by on the road. The image recognition techniques may fail due to poor lighting conditions and adverse environmental conditions. Further, the motorcycles can be tested for unlawful mixture of fuel, over speed, excess emission of smoke and the like, based on the sound. We have carried out the literature survey to know the state-of-the-art and the following works are cited. The time-domain and frequency-domain features of the acoustic signatures are used as input for the backpropagation (BP) neural network for classifying the motorcycles into bikes and scooters [1]. Signal energy, energy entropy, ZCR, spectral roll-off, spectral centroid (SC) and spectral flux are extracted from the vehicle sounds and used for detection and classification of moving vehicles [2]. Two feature extraction methods are investigated for acoustic signal-based classification of moving ground vehicles [3]. The first one is based on spectrum distribution and the second one on wavelet packet transform. The performances of k-nearest neighbour (k-NN) algorithm and support vector machine (SVM) are evaluated. A vehicle sound classification system is presented that uses time-encoded signal processing and recognition method combined with the archetypes technique [4]. The work implements different Butterworth low-pass filters for low-pass filtering. A novel methodology for statistical modelling and classification of acoustic signals collected from a wireless sensor network is discussed [5]. One-dimensional (1D) wavelet decomposition of the acoustic signal is performed and the resulting sub-band coefficients are modelled using the alphastable distribution. The similarity between two acoustic signals is measured by employing a variant of the Kullback–Leibler divergence between IET Intell. Transp. Syst., 2012, Vol. 6, Iss. 3, pp. 235–242 235 doi: 10.1049/iet-its.2011.0162 & The Institution of Engineering and Technology 2012 www.ietdl.org