Title

Pose estimation for soccer robot vision using genetic algorithms

Date of Publication

2002

Document Type

Master's Thesis

Degree Name

Master of Science in Computer Science

Subject Categories

Computer Sciences

College

College of Computer Studies

Department/Unit

Computer Science

Thesis Adviser

Elmer P. Dadios

Defense Panel Chair

Florante R. Salvador

Defense Panel Member

Eufemio Barcelon
Philip Chan

Abstract/Summary

This is an investigation of the applicability of a learning mechanism for the pose estimation of the soccer robot system used in Micro-Robot World Cup Soccer Tournament (MIROSOT). Current vision systems in a MIROSOT game use various techniques, such as conventional method and the use of specialized hardware for pose estimation. This research focused on pose estimation by incorporating genetic algorithm (GA) to the vision system. Two GAs were implemented: PGA/OGA (Position Genetic Algorithm/Orientation Genetic Algorithm) and GA2 (implementation of PGA/OGA combined). PGA/OGA performed the pose estimation task by subdividing the task into two parts, position estimation and orientation estimation. The first GA cycle is the PGA that is responsible for position estimation and the second is the OGA that is responsible for orientation estimation. GA2, on the other hand, performs the pose estimation task using one GA cycle. For PGA, the fitness function used is Histogram Similarity. The PGA has a 90.00% accuracy in determining the position of an object. For the OGA and GA2, several fitness functions were created, namely Two Circles, Two Triangles, Color Index and CTC (combination of Two Circles, Two Triangles and Color Index fitness function). The orientation results of GA is 80.00% accurate if only one object is present on the scene and 36.00% accurate when more objects are present on the scene. In terms of overall accuracy, the Two Triangles fitness function produced the best results of 64.83% accuracy. A drawback, however, is that GA inherently consumes a large amount of processing time.

Abstract Format

html

Language

English

Format

Print

Accession Number

TG03321

Shelf Location

Archives, The Learning Commons, 12F Henry Sy Sr. Hall

Physical Description

1 v. (various foliations) ; 28 cm.

Keywords

Robot vision; Computer vision; Genetic algorithms; Combinatorial optimization

This document is currently not available here.

Share

COinS