Identification of winter road friction coefficient based on multi-task distillation attention network

Feilin Liu, Yan Wu, Xinneng Yang, Yujian Mo, Yujun Liao

Abstract: Road friction coefficient estimation is an important task in the perception system of autonomous driving vehicles. It enables the vehicle to perceive upcoming road friction conditions and helps the decision-making system to adjust the driving styles accordingly in case of potential traffic accidents caused by tire slip. However, to our knowledge, there is currently no recognized image benchmark dataset in this field with enough weather diversity for this task. And many existing methods are measured under different standards. In this work, we present a road friction coefficient estimation dataset that includes all-weather traffic conditions, which is called the winter road friction (WRF) dataset. Then, a novel friction coefficient estimation model based on multi-task distillation attention network (MDAN) is proposed to solve this task in an end-to-end way. The proposed model surpasses existing methods in this field and reaches 86.53% Acc on the WRF dataset. The WRF dataset will be made publicly available at https://github.com/blackholeLFL/The-WRF-dataset soon.