Development of an anomalous noise event detection algorithm for dynamic road traffic noise mapping

Dynamic road traffic noise maps should display, in real time, the noise levels generated by road infrastructures measured by the sensors located on the road. For this reason, any acoustic event produced by another source that could alter the measured noise levels (e.g. an aircraft flying over, nearby railways, church bells, crickets, etc.) should be detected and eliminated from the map computation to provide a reliable picture of the actual road noise impact. To that end, it becomes necessary to devise strategies to automatically identify anomalous noise events captured by the network of sensors. This work describes a first version of the anomalous noise event detection algorithm designed in the LIFE DYNAMAP project. The proposed algorithm follows a "detection-by-classification" approach based on a semisupervised two-class classifier that does not require training with on-site collected "anomalous noise events" samples, thus being location-independent. Instead, it optimizes a decision threshold based on distance distributions with respect to the predominant "road traffic noise" class to maximize detection accuracy. The experimental results reveal that our proposal outperforms the baseline two-class supervised detector especially in scenarios in which anomalous events show higher noise levels and, thus, are more likely to alter the levels represented in dynamic road traffic noise maps.