Car passing audio dataset for traffic monitoring

Road traffic monitoring has long been an important task to ensure road quality, safety, and proper traffic control. In this paper, we present an acoustic-based approach and a corresponding dataset to tackle the relevant challenges in remote traffic monitoring such as vehicle counting, direction, and speed estimation. Our main contribution is the collection of a big dataset containing sound samples of cars passing by a constellation of six synchronized recording devices. The dataset consists of approx. 82500 separate acoustic samples that are related to 19000 car-passing events. We demonstrate in the paper the possible tasks that could be solved with the help of the data and implemented a wrapper Python class to help fellow researchers handle the dataset, which is publicly available. To validate our approach and the relevance of the dataset, we present promising preliminary results from classical ML approaches and neural network-based classifiers trained for traffic volume estimation and car sound detection.