Effects of the Feature Extraction from Road Surface Image for Road Induced Noise Prediction using Artificial Intelligence 2019-01-1565

Next-generation vehicles driven by motor such as electric vehicles and fuel cell vehicles have no engine noise, so the balance of interior noise is different from the vehicles driven by conventional combustion engine. In particular, road induced noise tends to be conspicuous in the low to middle vehicle speed range, therefore, technological development to reduce it is important task. The purpose of this research is to estimate the frequency of the peak level of road induced noise from the signals of sensors adopted for automatic driving without newly adding sensors for utilizing the estimation result as a reference signal to reduce road induced noise by active control or semi active control. Using the monocular camera which is the simplest image sensor, the peak frequency of the road induced noise is estimated from the road surface image ahead of the vehicle by machine learning. The feature quantities are extracted using three image recognition techniques (HOG, CNN, autoencoder) from the pixel data of road surface images. From the feature quantities obtained by the above method, the frequency characteristics of the road induced noise were estimated using deep learning, which is generally considered to be high estimation accuracy as a machine learning method. For the eight types of road surface, we compared the road induced noise data measured by the actual vehicle and by estimated using deep learning. It was found that the estimation result using feature quantities of road surface is more accurate than that using only pixel data of road surface.