LAUSR

Abstract Traffic noise has large consequences on the appreciation of the living quality close to roads and leads to speech interference, sleep disturbances, and general annoyance. Furthermore, it impacts the economy as it increases the costs for noise abatement and decreases the value of properties, which are close to noisy road transport infrastructure. However, existing methods to measure tire/road noise lack in mobility, in the possibility of measuring comprehensively, and are expensive and sophisticated to use. We present a novel method and an exploitation model to predict different types of road surfaces based on tire cavity sound acquired under normal vehicle operation. With the information of the road surface, further parameters can be estimated, such as tire/road noise, tire/road friction or rolling resistance. Our method can be applied comprehensively and is inexpensive. In contrast to special measurement vehicles with laser profilometers, normal vehicles can be equipped with our measurement system and the data can be automatically analyzed with our method. Therefore, the road infrastructure can not only be monitored in fixed intervals of about one to four years but constantly. We tested our classifier with unseen data, which were not used for training. Our final classifier after post-processing has an average precision and recall of respectively 95.1 and 90.6% and a accuracy of 91.8%. The output of our model could be fused with other parameters such as weather data, to create a digital map to automatically identify noisy roads or dangerous spots with low tire road friction. This might be an important input for advanced driver assistance systems. Based on the detected road hazards, an optimal regulation of the vehicle speed could be achieved. For the civil engineering departments and road infrastructure operators, this tool could be used to perform a more efficient maintenance of the whole road infrastructure.