Real-time ground marking analysis for safe trajectories of autonomous mobile robots

Abstract

Scene understanding is widely linked to the perception of it. Lane detection and tracking are commonly used in the context of autonomous transportation to estimate the drivable area on marked road. Real-time, accurate and efficient analyses are particularly critical when mobile robots are considered. Moreover, safe trajectories are of importance for those robotic vehicles evolving other vehicles or humans interactions. This paper proposes an unsupervised and robust vision-based approach that provides Lane Marking Detection (LMD) to help the positioning system. The approach is applied in two different environments of AKKA Research projects: Link & Go, an autonomous car; and AIR-COBOT, a collaborative mobile robot which inspects aircrafts during maintenance operations. The autonomous car has to travel on roads and to stay in its ego-lane. The mobile robot cooperates with human and has to circulate in the airport and in hangars. In both contexts, painted lanes on the ground could help the navigation system. We demonstrate the benefits of our proposal through an evaluation of the proposed approach on real datasets with appropriate metrics.