Automatic classification of road vehicles considering their pass-by acoustic signature

In order to assess the impact of environmental noise on a community, it is essential to accurately describe the aspects and characteristics of the encountered noises. In this context, it would be of special interest to dispose of environmental noise monitoring stations capable of not only measuring the noise levels but also identifying the sources producing those levels. To offer such functionality, an algorithm to automatically recognize the noise events is required. According to previous works, distinguishing between road vehicle noise sources (i.e., light vehicles, heavy vehicles and motorbikes) is a particularly complex problem. This paper proposes a recognition scheme that takes into account the perceived characteristics of road vehicles pass-by, which may be divided into different phases: approaching, passing and receding. By taking independent decisions for the pass-by phases, the proposed recognition scheme is able to improve the recognition of road traffic vehicles with respect to a traditional recognition scheme, specifically in 7% for light vehicles and in 4% for heavy vehicles. The conducted listening tests reveal that the averaged recognition accuracy obtained by the proposed system is slightly better than human ability to recognize noise sources.