Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Predictive and adaptive maps for long-term visual navigation in changing environments

Lucie Halodova, Eliska Dvorakova, Filip Majer, Tomas Vintr, Oscar Martinez Mozos, Feras Dayoub, Tomas Krajnik

Abstract

In this paper, we compare different map management techniques for long-term visual navigation in changing environments. In this scenario, the navigation system needs to continuously update and refine its feature map in order to adapt to the environment appearance change. To achieve reliable long-term navigation, the map management techniques have to (i) select features useful for the current navigation task, (ii) remove features that are obsolete, (iii) and add new features from the current camera view to the map. We propose several map management strategies and evaluate their performance with regard to the robot localisation accuracy in long-term teach-and-repeat navigation. Our experiments, performed over three months, indicate that strategies which model cyclic changes of the environment appearance and predict which features are going to be visible at a particular time and location, outperform strategies which do not explicitly model the temporal evolution of the changes.

Predictive and adaptive maps for long-term visual navigation in changing environments

Abstract

In this paper, we compare different map management techniques for long-term visual navigation in changing environments. In this scenario, the navigation system needs to continuously update and refine its feature map in order to adapt to the environment appearance change. To achieve reliable long-term navigation, the map management techniques have to (i) select features useful for the current navigation task, (ii) remove features that are obsolete, (iii) and add new features from the current camera view to the map. We propose several map management strategies and evaluate their performance with regard to the robot localisation accuracy in long-term teach-and-repeat navigation. Our experiments, performed over three months, indicate that strategies which model cyclic changes of the environment appearance and predict which features are going to be visible at a particular time and location, outperform strategies which do not explicitly model the temporal evolution of the changes.
Paper Structure (21 sections, 4 figures)

This paper contains 21 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Visual teach-and-repeat navigation
  4. Map management
  5. Not matched feature
  6. Correctly matched features
  7. Incorrectly matched features
  8. Discarding map features
  9. Static map strategy
  10. Latest map strategy
  11. Aggressive map strategy
  12. Strict map strategy
  13. Summary map strategy
  14. Multiple map strategy
  15. Score-based, adaptive map strategies
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Feature matching between the map and the current robot view: The first image shows that the use of score-based map adaptation strategy results in more correspondences compared to the use of a static map, shown in the second image. The green lines represent correct matches, while red lines incorrect ones. The score-based map adaptation ensures a sufficient number correct correspondences due to using information from the static map augmented with features from later traversals.
  • Figure 2: Frequency Map Enhancement (FreMEn) for visual localisation: The observations of image feature visibility (centre,red) are transferred to the spectral domain (left). The most prominent components of the model (left,green) constitute an analytic expression (centre,bottom) that represents the probability of the feature being visible at a given time (green). This is used to predict the feature visibility at a time when the robot performs self-localisation (blue). For further details, see a video at https://youtu.be/Qw1kS_5zVwE, repository at fremen.uk or article krajnik2017fremen.
  • Figure 3: Representative images of the dataset collected. Columns correspond to the same locations, with top images obtained during the day and bottom images captured during night.
  • Figure 4: Probability of registration error being smaller than a given number of pixels for different map update strategies.