Estimation of Frontal Road Type Using Machine Learning and Ultrasonic Waves

Abstract

The amount of acceleration and deceleration can be optimized when a vehicle can predict the types of road surfaces in advance. Myriads of methods for predicting road surfaces have been proposed, but they required costly equipment or had poor prediction performance. This paper suggests a different method for predicting road surfaces by recognizing that each material has its unique acoustic impedance. By transmitting ultrasonic waves into road surfaces that a vehicle intends to analyze, the reflected ultrasonic signals from the surface can be classified by a model developed from machine-learning. To measure the effectiveness of the method in a real-world situation, several types of specimens were created, and different sets of data were acquired from each test. Furthermore, the data were obtained from different road surfaces to verify the effectiveness of the method in the real world.