Lane Geometry Reconstruction: Least Square Curve Fit Versus Kalman Filter

Abstract

Abstract This paper presents a process for down range road geometry reconstruction from measurements (e.g., using computer vision) of lane markings. Lane marker information from both near-field and far-field cameras is assumed. A perception range extension algorithm is proposed to solve the field of view problem which occurs with a single-far-field sensor. A steady state Kalman filter and a least square curve fitting scheme are developed for road geometry modeling. Simulations are conducted with these two separate road modeling schemes. Results are compared and discussed in terms of the road geometry reconstruction accuracy and the computational speed. These results show that the Kalman filter approach offers substantial advantages over least square curve fitting in steady state conditions. However, the least square fit yields a better geometry estimation in the transitions.