SPIL: A Shape Preserving Method for Improving Lane Width Accuracy of HD Maps

The characteristic of high accuracy of HD maps is critical to ensure safety for autonomous driving. However, this accuracy evaluation on entire data is mostly infeasible by the measurements of absolute and relative accuracy due to the fact that the ground truth data is not easily obtained over a large geographical range. This paper proposes an alternate method using lane with accuracy for its expected or designed value is available in most occasions. Furthermore, we identify optimization opportunities and design a shape preserving algorithm SPIL for improving the accuracy of lane width through the adjustment of endpoints of assignable boundary line segments. Finally experiments are performed on a real typical urban road dataset, and experimental results demonstrate that SPIL algorithm achieves above fifty percent accuracy performance improvement.