Evaluating the Impact of Lane Marking Quality on the Operation of Autonomous Vehicles

Abstract

The quality of lane markings is pivotal for safe operations and efficient trajectory generations of connected and autonomous vehicles (AVs). However, most studies are devoted to enhancing in-vehicle detection systems and ignore the impact of faulty lane markings. An instrumented vehicle was employed to mimic the data input of an AV and real-world trials were conducted on (1) live motorways; and (2) a controlled motorway facility. From the live motorway data, causal factors affecting computer vision lane detection and classification algorithms were examined, and an enhanced lane classification algorithm was developed to overcome the limitations posed by poor lane markings. In the controlled motorway facility, experiments to modify the physical appearance of the lane markings were conducted to further test the performance of the developed algorithm. The detection rates of the developed lane classification algorithm were compared with the lane departure warning (LDW) system already implemented in the vehicle. Findings revealed that the LDW system is accurate over 95% and 54% of the time when lanes are faded by 50% and 75% respectively. Further testing on the quality of the lane markings was carried out virtually in such a way that the experiments were replicated in a simulation environment to (1) identify lane marking conditions that can be reliably adopted for safe operations of AVs, (2) estimate the effect of adverse weather and lighting conditions on road markings detection, and (3) address localization issues for AVs. Simulation results show that poor lane markings have a significant negative impact on AV safety, especially in inclement weather and poor light conditions inducing an increase in conflicts and delays. This can be compensated for if more sophisticated sensors are employed in AVs, and the operators of road networks develop lane-based digital road maps.