Road Surface Recognition Based on Vision and Tire Noise

Abstract

Road surface recognition is an important research content of automated driving environment perception, which can provide the basis for decision-making and motion planning of automated driving. To improve the accuracy and robustness of road surface recognition, this paper proposes a road surface recognition scheme based on the fusion of machine vision and tire noise. Firstly, we extracted LBP features and deep learning features from images and used PCA algorithm to reduce the dimension of features. Then, the tire noise features were extracted by using statistics and MFCC algorithm. We fused image features and acoustic features to generate feature-level fusion information and realized the cross-modal fusion of sensors. Finally, the feature vectors of multiple time series were constructed into a feature matrix, and we used a convolutional neural network to train a road surface recognition classifier, which fully considers the context information of the road surface. Experimental results show that this method has an accuracy of 92.7% for road surface identification.