Spoon Detection And Classification

Spoon detection and classification
on conveyor belts using computer
vision
Master thesis in cooperation with Jorgensen Engineering
A/S
Author: Lars Glud (290385)
Supervisor: Norbert Kr¨uger
Educational institution: University of Southern Denmark - SDU
Faculty of Engineering
Project start: 1. February 2010
Project delivery: 1. June 2010

Abstract
In connection with JorgensenEngineering A/S milk powder filling line, a
robot picks up plastic spoons from a conveyor belt and drop them into cans
one at a time. Unfortunately the robots vision system has trouble finding
some of the spoons which can be semi-transparent.
Given a camera, see figure 1, taking gray-scale images of the spoons on
a conveyor belt, this thesis describe the development of a vision algorithm
that makesit possible to detect transparent and colored spoons. The project
will concern detection of spoons that are clear of other spoons and not those
which are occluded by each other. When a possible spoon has been detected
it has to be classified either as a spoon or as a debris1. If the possible spoon
is classified as a spoon, the pose of the spoon has to determined and its
orientation marks it asa left, middle or right spoon.
By using ordinary image operators it was possible to develop a vision
algorithm that can detect all colored spoons and the majority of the transparent
spoons. In order to classify the possible spoons, three classifiers was
trained and compared to each other. The best classifier was found to be the
Support Vector Machine and it had a positive classification rateof 99.5979
% for the colored spoons and 99.7617 for the transparent spoons. Overall
the development of the algorithm was a success and it has potential to be
used in the business world.
1Debris refers to broken spoons, spoons occluded by each other or spoons in the image
boundary where only a part of the spoon is visible
3
May 31, 2010
Figure 1: Milk powder filling line robot
4 of 106Preface
As the last part of the education in Robotics a master thesis has to be
written which is divided into two parts. The first part is called FORK
which is prelemimary work to the master part, which is the second half.
The FORK part was used to develop a vision algorithm to detect spoons
and is the basis for the master project.
The master project is about classification of the spoonswhich is a requirement
from the cooperating company Jorgensen Engineering A/S which
made this project possible. The available hardware for the project was supplied
from Jorgensen Engineering A/S. Due to the cooperation with the
firm Jorgensen Engineering A/S, it has been of great importance that there
was an actual product in the end of the master, rather than a profound
investigation of how tosolve the problem.
Jorgensen Engineering A/S has functioned as an extern supervisor during
the project and this has in practice been meetings at Jorgensen Engineering
A/S and communication through other media such as phones and email.
A great thanks should therefore be given to Mr. Arne Madsen and Mr.
Karsten Rasmussen from Jorgensen Engineering A/S for the opportunity to
write the projectin collaboration with them and the supervision they have
given throughout the project.
A great thanks to my primary supervisor Mr. Norbert Kr¨uger who
has suggested how to approach the problem and guided me throughout the
project.
Lastly a great thanks to my fellow student Mr. Thomas Mosgaard Giselsson
who has been a great correspondence during the project work.
5

Contents
1 Introduction15
1.1 Motivation for the project . . . . . . . . . . . . . . . . . . . . 15
1.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3 Reading guide . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Key terms and concepts . . . . . . . . . . . . . . . . . . . . . 18
1.4.1 Digital image processing . . . . . . . . . . . . . . . . . 18
1.4.2 Features . . . . . ....