Hybrid Metric-topological Mapping for Large Scale Monocular SLAM

Abstract

Simultaneous Localization and Mapping (SLAM) is a central problem for autonomous mobile robotics. Monocular SLAM is one of the ways to tackle the problem, where the only input information are the images from a moving camera. Current approaches for this problem have achieved a good balance between accuracy and density of the map, however, they are not suited for large scale. In this paper, we present a dynamic mapping strategy where the metric map is divided into regions with highly connected observations, resulting in a topological structure which permits the efficient augmentation and optimization of the map. For that, a graph representation where the nodes represent keyframes, and their connections are a measure of their overlapping, is continuously rearranged. The experiments show that this hybrid metric-topological approach outperforms the efficiency and scalability of previous approaches.

Keyframe-based monocular SLAM: design, survey, and future directions

“…The conversion from a topological to a metric map is not always trivial, and for this reason, recent monocular SLAM systems [37,38,39,40] employ hybrid maps, which are locally metric and globally topological. The implementation of a hybrid map representation permits the system to first reason about the world at a high level, which allows for efficient solutions to loop closures and failure recovery using topological information; and second, to increase efficiency of the metric pose estimate, by limiting the scope of the map to a local region surrounding the camera [41]. A hybrid map allows for local optimization of the metric map, while maintaining scalability of the optimization over the global topological map [42].…”

A review of visual SLAM for robotics: evolution, properties, and future applications

“…The system workflow consists of four sequential stages ( Klien and Murray. 2007 ; Fernández-Moral et al, 2013 ). Input preparation and system initialization involve processes such as monocular camera translation and rotation to improve image efficiency and clarity ( De Croce et al, 2019 ).…”

“…It is designed for large-scale mapping jobs where it can create a very accurate and detailed map of the working fields. In addition, it stays accurate even with a lower image resolution ( Engel et al, 2015 ; Fernández-Moral et al, 2013 ). This flexibility makes it a better choice for operating in complex, wide-ranging and dynamic environments and is used in various applications such as robotics and self-driving cars ( Mur-Artal et al, 2015 ; Eng et al, 2014 ); see Table 2 .…”

“…Various types of maps can be generated using SLAM, including topological maps, volumetric (3D) maps, such as point cloud and occupancy grid maps, and feature-based or landmark maps. The choice of the map type is based on factors such as the sensors employed, application requirements, environmental assumptions, and the type of dataset used in robotic applications ( Taheri and Xia, 2021 ; Fernández-Moral et al, 2013 ). In robotics, a grid map is a representation of a physical environment, with each cell representing a particular location and storing data comprising obstacles, topography, and occupancy.…”

Uncertainty-based localization in a topological robot navigation system

“…The next works built a topological map from a global metric map and then partitioned the metric map into local metric maps [22,62,84,129,138]. Local metric maps offered a more tractable detailed representation of the environment that leaded to more efficient map building and path planning.…”