Low-Cost Surface Classification System Supported by Deep Neural Models

Abstract

Determining the surface on which a vehicle is moving is vital information for im-proving active safety systems. Performing the surface classification or estimating adherence through tire slippage can lead to late action in possible risk situations. Currently, approaches based on image, sound, or vibration analysis are emerging as a viable alternative, though sometimes complex. This work proposes a methodology based on the use of low-cost accelerometers combined with Deep Learning tech-niques. The performance of the proposed system is evaluated with real tests, where high percentages of accuracy are obtained in the classification task.