Road state estimation based on acoustic analysis

Abstract

An electronic system capable of estimating the road surface status in real time, and among dry/wet/icy/snowy is shown. The system is mainly based on the analysis of the tyre/road noise generated during vehicle driving. The sound emission pattern of the tyre/road interaction changes dramatically depending on whether the pavement is dry, wet, icy or snowy. Thus discriminating the tyre/road noise acoustic footprint, it is possible to estimate the road status. To be able to distinguish this acoustic footprint, a Support Vector Machines (SVM) based classifier has been developed. Tyre/road noise is initially captured using a properly conditioned microphone. The obtained signal is digitized and pre-processed, extracting its spectral components to produce the feature vector that will be fed to classifier. Finally the classifier estimates the condition of the asphalt, which may be broadcast through the vehicle communication interface, to be shown on the console, or to be used by other subsystems in order to improve safety and comfort. The connection to the vehicle engine control units through its communications interface, also allows obtaining information on its dynamic variables, allowing improving the results obtained by the classifier. Before the system is able to operate properly, the classifier needs to be trained. In initial tests, the system has been trained to distinguish the states of dry and wet asphalt, yielding success rates around 91 % and an average response time in the transition from dry to wet of about 0.2 s. These results demonstrate the feasibility of the system. Currently, a reference implementation of the system is under development. It integrates all the elements needed in a board of reduced size and low cost. Current implementation of the system could improve security by issuing a slippery asphalt warning, when it detects wet asphalt. Data obtained may also be used by the traction control subsystem to improve safety and comfort. Additional work is being made to extend the classifier algorithms to detect icy and snowy asphalt states.