2022 Volume 3 Issue J2 Pages 642-649
This paper proposes a novel method for estimating road surface conditions in winter by integrating data from multiple in-vehicle sensors. The proposed method is a multimodal model consisting of multiple discriminators that estimate road surface conditions from a camera, in-tire accelerometer, road surface thermometer, and microphone, and a discriminator that combines their outputs, road surface condition probabilities. The road surfaces to be estimated are the six road surfaces used in road management: dry, slightlywet, wet, slushy, icy, and snowy. The proposed method has been validated by utilizing actual data which are obtained from real-vehicle experiments on public winter roads, and its accuracy is shown to be superior to that of conventional methods.
