Construction process of a three-dimensional roadway geometry map for autonomous driving

Abstract

This paper presents a construction process of the three-dimensional roadway geometry map for an autonomous driving system. The presented process focuses on the post-processing and the three-dimensional roadway geometry modelling algorithms. The post-processing algorithm refines the raw Global Positioning System position data by combining the novel three-dimensional roadway geometry model which originated from the construction engineering of roadways and vehicle motion data from onboard sensors using a Rauch-Tung-Striebel smoother. The three-dimensional geometry modelling algorithm approximates the road geometry information described in three-dimensional point clouds into a mathematical curve model based on the B-spline. An adaptive curve refinement method using dominant points was applied to the road modelling algorithm. This dominant-point-based refinement method can reduce the number of knots and the number of control points of the B-spline road model while maintaining the desired accuracy of the roadway map. Also, since the dominant-point refinement method considers a road shape factor, such as the curvature and the arc length, for the road modelling, it is more efficient than the previous B-spline road modelling algorithms. The proposed map generation algorithm was verified and evaluated through experiments in various test conditions. The experimental results show that the presented construction process of the three-dimensional roadway geometry map can provide sufficient accuracy, reliability and efficiency for applications of autonomous driving systems in comparison with those of other roadway map construction processes.