Subpaving-Map: Fast Terrain Mapping for Outdoor Scenes based on Set-membership Theory

Abstract

Three-dimension terrain modeling of outdoor scenes mostly relies on probability theory. However, the lack of inherent robustness in probability methods can lead to the coupling and accumulation of uncertainties from multiple sources, making it difficult to ensure the reliability of the constructed terrain model. In order to build an accurate and highly robust dense terrain map, three-dimension terrain mapping based on set-membership theory is studied. Subpaving-Map, a fast terrain mapping method for outdoor scenes is proposed, which includes three modules: measurement modeling, feature association and terrain map update. Subpaving-Map fully considers the strict constraint boundary of terrain uncertainty during the process of mapping. Subpaving-Map is compared with Octomap on the KITTI dataset, and it is demonstrated that Subpaving-Map achieves a balance in accuracy, real-time performance and robustness.