Lane Geometry Perception and the Characterization of Its Associated Uncertainty

Abstract

This paper addresses the reconstruction of down-range road geometry from imaging sensors for application to motor vehicle active safety systems. This study assumes measurements of lane marker locations in the previewed scene are available from an imaging sensor. An algorithm is developed to extend the perception range of a single-far-field sensor to alleviate the field of view problem. Two steady-state Kalman filters and a least square curve fitting scheme are developed to compute estimates of the road geometry. Simulations are used to compare the performance of the different road modeling schemes for different roadway scenarios, providing insights useful for selecting model-based road geometry estimation techniques. Finally, an algorithm to characterize the uncertainty in road geometry perception is proposed and validated.