M2T: Local map descriptor

Abstract

Map matching, the ability to match a local map built by a mobile robot to previously built maps, is crucial in many robotic mapping, self-localization, and simultaneous localization and mapping (SLAM) applications. In this paper, we propose a solution to the “map-to-text (M2T)” problem, which involves the generation of text descriptions of local map content based on scene understanding to facilitate fast succinct text-based map matching. Unlike previous local feature approaches that trade discriminativity for viewpoint invariance, we develop a holistic view descriptor that is view-dependent and highly discriminative. Our approach is inspired by two independent observations: (1) The behavior of mobile robots given a local map can often be characterized by a unique viewpoint trajectory, and (2) a holistic view descriptor can be highly discriminative if the viewpoint is unique given the local map. Our method consists of three distinct steps: (1) First, an informative local map of the robot's local surroundings is built. (2) Next, a unique viewpoint trajectory is planned in accordance with the given local map. (3) Finally, a synthetic view is described at the designated viewpoint. Because the success of our holistic view descriptor depends on the assumption that the viewpoint is unique given a local map, we also address the issue of viewpoint planning and present a solution that provides similar views for similar local maps. Consequently, we also propose a practical map-matching framework that combines the advantages of the fast succinct bag-of-words technique and the highly discriminative M2T holistic view descriptor. The results of experiments conducted using the publicly available radish dataset verify the efficacy of our proposed approach. Further, although this paper focuses on the standard 2D pointset map, we believe that our approach is sufficiently general to be applicable to a broad range of map formats, such as the 3D and general view-based maps.