Visual Topological Mapping Using an Appearance-Based Location Selection Method

Authors Organisations
Type Conference Proceeding (Non-Journal item)
Original languageEnglish
Title of host publicationTowards Autonomous Robotic Systems - 21st Annual Conference, TAROS 2020, Proceedings
EditorsAbdelkhalick Mohammad, Xin Dong, Matteo Russo
PublisherSpringer Nature
Number of pages13
ISBN (Electronic)9783030634865
ISBN (Print)9783030634858
Publication statusE-pub ahead of print - 03 Dec 2020
Event21th Annual Conference on Towards Autonomous Robotics, TAROS 20120 - Nottingham, United Kingdom of Great Britain and Northern Ireland
Duration: 16 Sep 202016 Sep 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12228 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference21th Annual Conference on Towards Autonomous Robotics, TAROS 20120
Country/TerritoryUnited Kingdom of Great Britain and Northern Ireland
Period16 Sep 202016 Sep 2020
Permanent link
Show download statistics
View graph of relations
Citation formats


Visual representation of an environment in topological maps is a challenging task since different factors such as variable lighting conditions, viewpoints, mobility of robots, dynamic and featureless appearance, etc., can affect the representation. This paper presents a novel method for appearance-based visual topological mapping using low resolution omni-directional images. The proposed method employs a pixel-by-pixel comparison strategy. Successive images captured as a mobile robot traverses its environment are compared to estimate their dissimilarity from a reference image. Specific locations (nodes in the topological map) are then selected using a variable sampling rate based on changes in the appearance of the environment. Loop-closures are created using a dynamic threshold based on variability of the environment appearance. The method therefore proposes a full SLAM solution to create topological maps. The method was tested on multiple datasets, which were captured under different weather conditions along various trajectories. GPS coordinates were used to stamp each image as ground truth for evaluation and visualisation only. We also compared our method with state of the art feature-based methods.


  • Appearance-based, Loop closure, Topological mapping, Visual SLAM