|
Abstract
|
BACKGROUND:
Faced with the high rate of traffic accidents under slippery road conditions, agencies attempt to quickly identify slippery spots on the road and drivers want to receive information on the impending dangerous slippery spot, also known as "black ice."
Methods:
In this study, wheel slip, defined as the difference between both speeds of vehicular transition and wheel rotation, was used to detect road slipperiness. Three types of experiment cars were repeatedly driven on snowy and dry surfaces to obtain wheel slip data. Three approaches, including regression analysis, support vector machine (SVM), and deep learning, were explored to categorize into two states-slippery or non-slippery.
Results:
Results indicated that a deep learning model resulted in the best performance with accuracy of 0.972, only where sufficient data were obtained. SVM models universally showed good performance, with average accuracy of 0.965, regardless of sample size.
Conclusion:
The proposed models can be applied to any connected devices including digital tachographs and on-board units for cooperative ITS projects that gather wheel and transition speeds of a moving vehicle to enhance road safety in winter season though collecting followed by providing dangerous slippery spots on the road.
Keywords: Wheel slip , Road slipperiness , Connected vehicle , Support vector machine , Deep leraning , Regression analysis.
Language: en |