Towards Perspective-Free Pavement Distress Detection via Deep Learning

This paper introduces a pavement distress detection framework based on a devised deep convolutional neural network for improving the automatic driving safety on intelligent vehicles. In contrast to the conventional detection methods by utilizing multiple sensors, the proposed framework directly profiles the features of road conditions from the incoming road images as the perspective maps. Experiment results demonstrate that the proposed framework achieves superior representation of pavement distress with higher accuracy than that made by the baseline convolutional neural network.