Global and local frameworks for vehicle state estimation using temporally previewed mapped lane features

Abstract

This paper proposes a method for using a forward-looking monocular camera along with previewed road geometry from a high-fidelity, low-dimensional map to estimate lateral planar vehicle states by measuring the vehicle's temporally anticipated reference trajectory. Theoretical estimator performance from a steady-state Kalman Filter implementation of the estimation framework is calculated for various look-ahead distances and vehicle speeds. Application of this filter structure to real driving data is also briefly discussed. The use of temporally previewed measurements of a vehicle's reference path is shown to greatly improve the accuracy of vehicle planar state estimates, and shows promise for use in closed-loop lane keeping and driver assist applications.