Reliable Identification of Road Surface Condition Considering Shadow Interference

Abstract

In order to achieve precise motion control of the vehicle under various road surface conditions, it is essential to identify the road surface conditions accurately and quickly. For low-cost cameras and computing platforms, this paper proposes a reliable data-level fusion identification method that is robust to scenes with shadow. Based on deep-learning models, the candidate regions of interest with confidence level for each road surface condition are obtained. Besides, the fusion identification method based on improved Dempster-Shafer Evidence Theory is designed. The results of evaluation on specific dataset with shadow and anti-lock braking system using proposed method reveal that our proposed fusion method achieves superior performance with stable and accurate predictions.