Simple Search:

Topological mapping and qualitative localization based on K-adjacent union clustering algorithm


Karasfi, Babak (2012) Topological mapping and qualitative localization based on K-adjacent union clustering algorithm. PhD thesis, Universiti Putra Malaysia.

Abstract / Synopsis

In robotic applications, localization and mapping as parts of the navigation system are fundamental competence for mobile autonomous systems. The position of the mobile robot is known as qualitative localization inside a topological map, where the place recognition is an essential problem to overcome. Previously, supervised place recognition approaches have been used to solve global localization in offline mode. The aim of this thesis is to develop a mobile robot topological mapping and qualitative localization method based on unsupervised and fully appearance-based place recognition approach. In this research, two different methods are designed and implemented to answer the aim of this thesis. These methods focus on perspective or omnidirectional image similarity based on local features or the combination of global and local features which are identified as speed-up robust feature (SURF) and hue saturation intensity (HIS) color histogram. Moreover, proposed methods are spatial and sequential based place clustering methods (unsupervised learning) which are try to find the representative image that is more similar to the current adjacent robots query image. Therefore, the topological map graph of the place clusters can be created and qualitative localization can be performed over the topological map graph. According to the experimental results, the average of recognition precision for the first offline proposed method is 95% and in a different illumination condition is 86%. Moreover this performance in the kidnapped robot experiment is more than 90%. The average of online place recognition percentage for the second online, incremental and expandable proposed method is 93.56% and in different illumination conditions is 86.06%. In addition, the average performance of the topological mapping and qualitative localization results, obtained from expanded-environment experiments is 91.71%. Considering all results, the proposed topological mapping and qualitative localization methods are robust, accurate, cost effective, portable, low power consumption, low weight, easy to install without any camera calibration and can be applied on various mobile robot platforms.

Download File

ITMA 2012 12R.pdf

Download (947kB) | Preview

Additional Metadata

Item Type: Thesis (PhD)
Subject: Topographic maps
Subject: Cluster analysis
Subject: Algorithms
Call Number: ITMA 2012 12
Chairman Supervisor: Associate Professor Tang Sai Hong, PhD
Divisions: Institute of Advanced Technology
Depositing User: Haridan Mohd Jais
Date Deposited: 24 Feb 2015 13:54
Last Modified: 24 Feb 2015 13:54
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item