Terrain Traversability Mapping Based on LiDAR and Camera Fusion

Abstract

Terrain traversability mapping plays an important role in autonomous exploration of unmanned ground vehicles. In many cases, information from a single sensor such as LiDAR or camera may not be sufficient for estimating traversability reliably. For example, LiDAR-based methods are better at identifying areas with strong structural characteristics, rather than cluttered areas, such as lawns. Vision-based methods can distinguish different regions with semantic meanings. However, sometimes there may be a misclassification due to a domain gap or other reasons, which will make it risky during the robot’s navigation process. In this work, we propose a novel LiDAR-vision-based method for terrain traversability mapping. Our method is mainly composed of three modules: vision-based traversable area segmentation, LiDAR-based traversable area extraction, and Bayesian fusion. Experimental results demonstrate that the proposed method is able to fulfill real-time and reliable traversability mapping and shows superior to the state-of-the-art method.