A support vector machine approach to a classification problem in robotics

Wang, Jianxiong. (2004). A support vector machine approach to a classification problem in robotics MPhil Thesis, School of Information Technology and Electrical Engineering, The University of Queensland.

       
Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
THE17886.pdf Full text application/pdf 2.43MB 2
Author Wang, Jianxiong.
Thesis Title A support vector machine approach to a classification problem in robotics
School, Centre or Institute School of Information Technology and Electrical Engineering
Institution The University of Queensland
Publication date 2004
Thesis type MPhil Thesis
Supervisor Professor Tom Downs
Dr. Gordon Wyeth
Total pages 54
Language eng
Subjects 290901 Electrical Engineering
Formatted abstract The problem of simultaneous localization and mapping (SLAM) has been the subject of widespread study in the field of robotics in recent years. Researchers into SLAM are aiming to incrementally build a map of the environment while concurrently using this map to compute robot location. An efficient solution to the problem depends on accurate measurement of the environment which in turn requires a robust and quick method to amalgamate data from various sensors.

This thesis is concerned with one component of the SLAM problem, namely the problem of object recognition. For this problem the robot we employ has an array of sonar sensors and these provide the data we employ for the object recognition task. The sonar data is preprocessed and then fed to a support vector machine (SVM) classifier to carry out the recognition.

The thesis describes the use of genetic algorithms (GAs) to assist the SVM recognition process in two novel ways. It is shown how a GA can be employed firstly to optimize the tuning parameters of an SVM classifier and secondly to select a sequence of subsets of the training data so that the generalization performance of the SVM gradually improves.
Keyword Robotics

 
Citation counts: Google Scholar Search Google Scholar
Access Statistics: 101 Abstract Views, 3 File Downloads  -  Detailed Statistics
Created: Mon, 03 Oct 2011, 12:13:58 EST by Miss Wendy James on behalf of Scholarly Publishing and Digitisation Service