Speed-dependent wet road surface detection using acoustic measurements, octave-band frequency analysis and machine learning algorithms

Stratis Kanarachos, Mike Blundell, Michalis Kalliris, Rigas Kotsakis

Research output: Chapter in Book/Report/Conference proceedingConference proceeding

Abstract

Rain, even shortfall rain, can adversely influence road safety. Currently, most methods for the detection of slippery road conditions require considerable excitation – by braking or steering – of the tire forces. Thus, the safety margin between detection and reaction is minimal and dangerous situations can occur, for example hydroplaning. In this study, we explore the detection of wet road conditions by utilizsing a relatively inexpensive technology, acoustic measurements. An experimental investigation took place at Coventry, UK. A vehicle instrumented with a microphone, GPS and CAN bus data logger was driven at a wide range of speeds and for different road wetness conditions. A wet road surface detector for high and low speeds was developed and evaluated. The detector included a ground-truth database that was constructed by combining the collected speed data, the road surface condition and the audio features that were calculated using octave-band frequency analysis. The classification results by testing the performances of different machine learning algorithms on the database are discussed. The best results were obtained with Support Vector Machines. Future research directions are drawn.
Original languageEnglish
Title of host publicationISMA conference on Noise and Vibration Engineering
Publication statusPublished - 18 Sep 2018
EventISMA conference on Noise and Vibration Engineering - Leuven , Belgium
Duration: 17 Sep 201819 Sep 2018
http://past.isma-isaac.be/isma2018/

Conference

ConferenceISMA conference on Noise and Vibration Engineering
CountryBelgium
City Leuven
Period17/09/1819/09/18
Internet address

Fingerprint Dive into the research topics of 'Speed-dependent wet road surface detection using acoustic measurements, octave-band frequency analysis and machine learning algorithms'. Together they form a unique fingerprint.

Cite this