Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Keyframe-based Dense Mapping with the Graph of View-Dependent Local Maps

Krzysztof Zielinski, Dominik Belter

TL;DR

This work tackles dense mapping for static environments by replacing uniform 3D voxel-based representations with a 2D view-dependent scheme of local maps built from keyframes. Each keyframe maintains a 2D container of ellipsoids that encode 3D occupancy with covariances, updated from RGB-D data via an NDT-inspired update, and organized in a pose graph to support loop closures and global map correction. The approach includes ellipsoid filtering, cross-map merging with clustering, and occlusion handling to produce a coherent global map with varying resolution that emphasizes near-field detail. Compared to OctoMap and NDT-OM, the method achieves higher detail for close objects, reduces storage through clustering, and enables efficient global map updates after loop closure, with practical implications for local planning and object detection in mobile robotic systems.

Abstract

In this article, we propose a new keyframe-based mapping system. The proposed method updates local Normal Distribution Transform maps (NDT) using data from an RGB-D sensor. The cells of the NDT are stored in 2D view-dependent structures to better utilize the properties and uncertainty model of RGB-D cameras. This method naturally represents an object closer to the camera origin with higher precision. The local maps are stored in the pose graph which allows correcting global map after loop closure detection. We also propose a procedure that allows merging and filtering local maps to obtain a global map of the environment. Finally, we compare our method with Octomap and NDT-OM and provide example applications of the proposed mapping method.

Keyframe-based Dense Mapping with the Graph of View-Dependent Local Maps

TL;DR

This work tackles dense mapping for static environments by replacing uniform 3D voxel-based representations with a 2D view-dependent scheme of local maps built from keyframes. Each keyframe maintains a 2D container of ellipsoids that encode 3D occupancy with covariances, updated from RGB-D data via an NDT-inspired update, and organized in a pose graph to support loop closures and global map correction. The approach includes ellipsoid filtering, cross-map merging with clustering, and occlusion handling to produce a coherent global map with varying resolution that emphasizes near-field detail. Compared to OctoMap and NDT-OM, the method achieves higher detail for close objects, reduces storage through clustering, and enables efficient global map updates after loop closure, with practical implications for local planning and object detection in mobile robotic systems.

Abstract

In this article, we propose a new keyframe-based mapping system. The proposed method updates local Normal Distribution Transform maps (NDT) using data from an RGB-D sensor. The cells of the NDT are stored in 2D view-dependent structures to better utilize the properties and uncertainty model of RGB-D cameras. This method naturally represents an object closer to the camera origin with higher precision. The local maps are stored in the pose graph which allows correcting global map after loop closure detection. We also propose a procedure that allows merging and filtering local maps to obtain a global map of the environment. Finally, we compare our method with Octomap and NDT-OM and provide example applications of the proposed mapping method.
Paper Structure (11 sections, 6 equations, 8 figures, 3 tables)

This paper contains 11 sections, 6 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Approach and Contribution
  4. View-dependent Mapping
  5. Global graph-based map
  6. Local maps - keyframes
  7. Update of the local map
  8. Ellipsoids filtering
  9. Merging local maps
  10. Results
  11. Conclusions and Future Work

Figures (8)

  • Figure 1: Local map of the environment: normal distribution transforms (ellipsoids) projected on the image plane.
  • Figure 2: Block diagram of the view-dependent keyframe-based mapping method
  • Figure 3: Example local map generated using, consecutively, 1, 2 and 10 frames, obtained for the freiburg1_desk sequence from the TUM dataset Sturm2012: 2D container with ellipsoids (a,b,c) and the enlarged region of the local map (d,e,f)
  • Figure 4: Merging two local maps (a,b): maps in the common coordinate frame (c) and the obtained set of ellipsoids (d).
  • Figure 5: Occluding ellipsoids removal procedure which allows keeping ellipsoids which represent objects with the highest precision: before (a) and after filtering (b)
  • ...and 3 more figures