A random finite set approach to multiple lane detection

Robust lane detection is the precondition for advanced driver assistance systems like lane departure warning and overtaking assistants. While detecting the vehicle's lane is sufficient for lane departure warning, overtaking assistants or autonomous driving functions also need to detect adjacent lanes. In this contribution, a novel approach to multiple lane detection based on multi-object Bayes filtering is presented. This method allows for directly considering the dependencies between multiple lanes without explicit data association in post processing. Furthermore, the proposed lane detection algorithm is applied to a challenging scenario of a rural road.

I. INTRODUCTION

In recent years, lane departure warning systems have become available on the market of automotive driver assistance systems. In order to warn the driver of an unintended passing of the lane, these systems have to be able to detect the lane markings just before the car is about to pass them. In the case of more advanced driver assistance systems, such as an overtaking assistant, lane markings have to be detected in much greater distance in front of the car. Furthermore, not only the lane markings of the lane the vehicle is currently driving on have to be detected, but also the markings of the adjacent lanes.