Stochastic road shape estimation

Abstract

We describe a new system for estimating road shape ahead of a vehicle for the purpose of driver assistance. The method utilises a single on board colour camera, together with inertial and velocity information, to estimate both the position of the host car with respect to the lane it is following and also the width and curvature of the lane ahead at distances of up to 80 metres. The system's image processing extracts a variety of different styles of lane markings from road imagery, and is able to compensate for a range of lighting conditions. Road shape and car position are estimated using a particle filter. The system, which runs at 10.5 frames per second, has been applied with some success to several hours' worth of data captured from highways under varying imaging conditions.