A novel lane support framework for vision-based vehicle guidance

The aim of this paper is the development of a robust lane support system using images from a single camera mounted on the ego-vehicle. The road boundary is tracked from the near to the far region of the image, by fusing information from the previous road models, the edge pixels selected in the current image, as well as edge pixels detected at the current image position of the tracking. The road boundaries are modelled by explicit polynomials, whose order is determined online. The prediction of the road boundaries using both a constrained state space model and an ARMA model is investigated. Finally, using the derived road models, a state space model estimates the parameters required for vehicle navigation.