Localization Based on Floor Line Map Using Monocular Camera for Factory Autonomous Mobile Robot

This study presents a localization for factory autonomous mobile robot based on a floor white line map with transforming the line recognition results from a monocular camera mounted on the front of the robot into a bird’s-eye view. The Monte Carlo Localization with particles based on the observation of floor lines is used. However, the projection of white lines onto the floor coordinates through the bird’s-eye view transformation is affected by variations in the robot’s pitch angle and observational noise caused by reflections from the floor lights. Especially when white lines are not sufficiently observed, updating the likelihood can lead particles to converge to incorrect positions, resulting in inaccurate localization. In this study, we propose a method that updates the likelihood only when a sufficient amount of observations is available based on white line recognition results. We conducted real-world experiments in an actual factory environment with white line layouts and compared and discussed the proposed method with a method using a LiDAR.