3D Lane Boundary Tracking Using Local Linear Segments

In this paper we propose a method for estimating arbitrary lane boundaries using noisy sensor data. Lane boundaries are modelled three dimensionally as consecutive linear segments in real world coordinates fixed in the local environment of the vehicle. A detailed error model is defined in order to estimate and represent uncertainties arising during the perception process. Uncertainties are estimated in lateral direction along lane boundaries. A detailed error model of the vehicle state incorporates the uncertainties arising during the perspective mapping. A two dimensional Interval Map is employed in order to structure the environment and manage the linear segments efficiently. The proposed method enables a compensation of erroneous ego state estimates or a flat world assumption. Results demonstrating the successful three dimensional estimation of lane boundary segments using real world sensor data are presented and discussed. The estimated positions are compared to reference data for evaluation.