On the quantitative assessment of the Lane Departure Warning System based on road scenes simulator

Abstract

Vision-based Lane Departure Warning Systems (LDWSs) have been studied for over two decays. This paper presents an Objective Evaluation Platform of LDWS (OEP-LDWS). It provides simulated road scenes with the possible data as ground truth, such as the vehicle to road relation, vehicle's states and the real Time-to-Lane-Crossing (TLC) value. In our OEP-LDWS, different kinds of driving maneuver can be simulated with the road model, the vehicle model, the camera model and a vehicle trajectory generator. At the same time, the road scene that may be captured by the on board camera can be generated. Using our OEP-LDWS, one can not only evaluate the warning performance of the LDWS quantitatively, but also assess the whole performance under varying circumstance, such as different road surfaces, different road curvatures and so on. Actually, those assessments can hardly be evaluated through real driving test, and are very important aspects for a LDWS, such as warning strategy selection, system tailor. Using our OEP-LDWS, we assess our LDWS with three different warning strategies, and reach the conclusion that the PTLC is the best under the low false warning criterion.