Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dominating Set Database Selection for Visual Place Recognition

Anastasiia Kornilova, Ivan Moskalenko, Timofei Pushkin, Fakhriddin Tojiboev, Rahim Tariverdizadeh, Gonzalo Ferrer

TL;DR

A novel approach for creating a visual place recognition database for localization in indoor environments from RGBD scanning sequences by utilizing a dominating set algorithm applied to a graph constructed from spatial information, referred to as the “DominatingSet” algorithm.

Abstract

This paper presents an approach for creating a visual place recognition (VPR) database for localization in indoor environments from RGBD scanning sequences. The proposed approach is formulated as a minimization problem in terms of dominating set algorithm for graph, constructed from spatial information, and referred as DominatingSet. Our algorithm shows better scene coverage in comparison to other methodologies that are used for database creation. Also, we demonstrate that using DominatingSet, a database size could be up to 250-1400 times smaller than the original scanning sequence while maintaining a recall rate of more than 80% on testing sequences. We evaluated our algorithm on 7-scenes and BundleFusion datasets and an additionally recorded sequence in a highly repetitive office setting. In addition, the database selection can produce weakly-supervised labels for fine-tuning neural place recognition algorithms to particular settings, improving even more their accuracy. The paper also presents a fully automated pipeline for VPR database creation from RGBD scanning sequences, as well as a set of metrics for VPR database evaluation. The code and released data are available on our web-page~ -- https://prime-slam.github.io/place-recognition-db/

Dominating Set Database Selection for Visual Place Recognition

TL;DR

A novel approach for creating a visual place recognition database for localization in indoor environments from RGBD scanning sequences by utilizing a dominating set algorithm applied to a graph constructed from spatial information, referred to as the “DominatingSet” algorithm.

Abstract

This paper presents an approach for creating a visual place recognition (VPR) database for localization in indoor environments from RGBD scanning sequences. The proposed approach is formulated as a minimization problem in terms of dominating set algorithm for graph, constructed from spatial information, and referred as DominatingSet. Our algorithm shows better scene coverage in comparison to other methodologies that are used for database creation. Also, we demonstrate that using DominatingSet, a database size could be up to 250-1400 times smaller than the original scanning sequence while maintaining a recall rate of more than 80% on testing sequences. We evaluated our algorithm on 7-scenes and BundleFusion datasets and an additionally recorded sequence in a highly repetitive office setting. In addition, the database selection can produce weakly-supervised labels for fine-tuning neural place recognition algorithms to particular settings, improving even more their accuracy. The paper also presents a fully automated pipeline for VPR database creation from RGBD scanning sequences, as well as a set of metrics for VPR database evaluation. The code and released data are available on our web-page~ -- https://prime-slam.github.io/place-recognition-db/
Paper Structure (12 sections, 6 equations, 3 figures, 3 tables)

This paper contains 12 sections, 6 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem formulation for optimal database
  5. Spatial overlap measure
  6. Overlap measure
  7. Graph processing
  8. Experimental results
  9. Datasets
  10. Database size and reduction rate
  11. Visual place recognition
  12. Conclusion

Figures (3)

  • Figure 1: Overview of the proposed methodology for building an optimal database for Visual Place Recognition (VPR). Top: a sequence of RGBD images obtained from the environment scanning. Bottom: (i) a 3D environment map is built from the scanning sequence, (ii) an overlap measure is estimated for each image pair using the spatial overlap in the voxelized map, (iii) the optimal VPR database is built by solving the dominating set problem for a graph where the localized images are vertices connected based on their estimated overlap, (iv) the rest of the images are split into database classes for VPR fine-tuning on the scanned scene.
  • Figure 2: Examples showcasing different overlaps between a query and a database image, interpreted as good or bad. Left: the database image covers a significant portion of the query image. Right: the query image is not well covered by the database image.
  • Figure 3: Examples of different scenes captured in Skoltech campus having similar structures. Left: the map of the Skoltech campus sequence. Right: pairs of images that are very similar visually but were captured in different locations.