Road course estimation using deep learning on radar data

Abstract

One of the most fundamental tasks in autonomous driving is the recognition of the road ahead. Using radar data, this is usually done via rule based algorithms. This paper proposes a deep learning approach to estimate the course of the ego lane based on occupancy grids generated by radar sensors. The method is also able to simultaneously give a reliability measurement of the predicted driving path. An automatic labeling process is engaged by utilizing the known ego pose of the vehicle obtained by a high precision positioning sensor. Due to its automated labeling process, learning data can be built up very cost efficiently.