Accurate vehicle pass-by noise emission quantification in real-life traffic

L’agence européenne pour l’environnement et l’organisation mondiale de la santé ont estimé que le bruit des transports est responsable de 100 000 morts prématurées par an en Europe. Le coût social du bruit en Europe est estimé à 40 milliards d’euro annuel. 90% de ce coût est relatif au trafic routier.

Le normes Européennes sont de plus en plus restrictives et ont forcé les constructeurs à produire des véhicules plus silencieux. Les véhicules dont la cartographie moteur ou le système d’échappement ont étés modifiés sont souvent légalement  réprimandables, mais en pratique restent complexes à appréhender. Ce papier présente le système innovant DBFlash, qui permet le monitoring acoustique pour la détection, l’identification et la verbalisation des véhicules bruyants. Le but étant une utilisation similaire aux radars de vitesse

Acoustic monitoring aims at detecting, identifying and classifying sound sources. Recently, applications such as street sound events detection or vulnerable areas protection have taken advantage of its techniques. Sound sources of interest can be as varied as dog barks, engine or drone noises and even gunshots sounds. Noises measurement and data processing are two key aspects of such applications. Noise measurement can be carried out with a single microphone but in most environment, the presence of both multiples sources and high background noise (wind, traffic) requires a microphone array to extract the signal of interest and localize the sound sources. This paper presents a methodology to control loud vehicle pass-by noise in streets. First, a specific noise camera, including acoustic and video measurements, allows to localise
the sound sources. Limited to one line of microphones for cost and compactness reasons, this array is not able to separate
the source of interest (a vehicle to control) from the background noise in its main directivity lobe. To do so, statistical learning is leveraged, using multimodal measurements to eliminate false alarms. This paper details the proposed joined methodology to localize and identify vehicle pass-by noise, and presents applications on experimental data.

Keywords: source localization, source identification, pass-by noise, acoustic monitoring