In recent years, a lot of effort has been put in vehicle safety systems for manned and unmanned drive. Road conditions are crucial among the factors that influence the choice of the driving style and the safety systems. In cooperation with ASK Industries we have developed a large dataset of car recordings employing acoustic sensors for the automated detection of road surface roughness and wetness. These have been used to develop deep learning algorithms able to obtain reliable outcomes in real-world conditions. Additionally, the detection of the road conditions allows novel in-car equalization and speech-enhancement scenarios.

This page reports data and publications related to this work-in-progress project.

Processing Acoustic Data With Siamese Neural Networks for Enhanced Road Roughness Classification (under review for IJCNN2019)

CNN and Siamese configuration files (zip)

How to read the configuration files: the CNN and Siamese configurations can be found under config_file_CNN and config_file_siamese, respectively. Under both folders you will find four folders ending with *_S_S, *_W_S and so on. This are the four training_testing configurations where S stands for summer tyres and W for winter tyres. The .conf files contained in each folder are numbered as in the paper and contain settings according to standard keras namings, e.g. the option

--dense-activation tanh

means that the activation function of the "dense" (fully connected) layers are of the tanh type.