Localization with Laser Range Finder in a Metrically Inconsistent Map from Monocular SLAM

Abstract

Due to the lack of a fixed baseline between frames, scale drift of the map from a monocular vision-based SLAM system is almost inevitable. It is difficult for a robot with range-based sensors to localize itself on such a metrically inconsistent map. In this paper, we propose a scale-aware localization method, which allows a robot with 2D LRF to achieve the purpose of localizing itself on a map from monocular SLAM. The scales of drifted local maps will also be corrected at the same time. In this work, we firstly present a technique to extract 2D structure from the 3D point cloud established by a monocular camera. And then, by employing an alternative to Particle Filter Localization algorithm, our approach is able to estimate both the poses of the robot and the local scales of the map. Finally, the performance of our approach has been evaluated on a public dataset.