Regresar

Detecting multiple simultaneous vehicles pass-by by sound source separation techniques

Abstract:

This paper presents new results on traffic noise classification applied to inter-city routes environments. The aim of the job is to improve the performance of an automatic classification system for traffic noise, which is able to detect, count and classify the traffic of a road once the signal has been recorded with a pair of microphones. Previous works showed the possibility to use automatic classification of traffic to define the number of vehicles and their classification. This information is critical to perform measurements of noise emission of a road according to the standard ISO 1996-2, and, of course it is relevant to apply the methods to calculate a noise map. The inter-class separation plays a relevant role to get a good classification. The transition from light to heavy vehicles is fuzzy, so there is a need increase the number of classes to improve the performance of the classifiers. An analysis of the optimal number of classes is required and, at the same time, a study of the temporal and spectral features to incorporate into the classification process it is also needed. Main algorithms of classification (FLD, k-NN and PCA) are compared and interesting results are extracted, concluding a better behavior for this last one in terms of rate of success and computation time.