Advanced

Applications of Machine Vision — Quality Control, Cancer Detection and Traffic Surveillance

Källén, Hanna LU (2016)
Abstract
During the last decades, image analysis has become an important tool in various applications. The increase in computer power has led to that more advances algorithms can be developed. In this thesis image analysis has been used in three different applications; quality control of asphalt, cancer detection in histopathological images and traffic surveillance.



In the first part of the thesis image analysis has been used to assist researchers at asphalt laboratories. Asphalt consists of a mixture of stones of different sizes and a binder called bitumen. Mainly two problems have been studied; the affinity between stones and bitumen and grain size distribution in asphalt samples. The affinity between bitumen and stone is very... (More)
During the last decades, image analysis has become an important tool in various applications. The increase in computer power has led to that more advances algorithms can be developed. In this thesis image analysis has been used in three different applications; quality control of asphalt, cancer detection in histopathological images and traffic surveillance.



In the first part of the thesis image analysis has been used to assist researchers at asphalt laboratories. Asphalt consists of a mixture of stones of different sizes and a binder called bitumen. Mainly two problems have been studied; the affinity between stones and bitumen and grain size distribution in asphalt samples. The affinity between bitumen and stone is very important for the durability of the pavement and is traditionally measured by the rolling bottle test. A goal in this thesis is to replace the manual evaluation in the method by automatic image analysis to make this test more accurate and objective. Another quality control is to estimate the size distribution of stones in asphalt samples to see if it follows the recipe for the asphalt. To avoid toxic substances, that today are used to dissolve the sample to retrieve the stones, we use image analysis to analyze cross sections of the samples.



In the second part, histopathological images have been studied. Features from pre-trained deep neural networks have been used to classify images of prostatic tissue. These features were used to train classifiers to classify the image into four different classed; benign tissue and three different grades of cancerous tissue. Doing this automatically will help the pathologist to speed up their diagnostics in the future.



The third and last part of the thesis deals with tracking and 3D-reconstruction of vehicles. By doing 3D-reconstructions of vehicles it is possible to accurate estimate their positions. A system to track interesting points between images in a movie sequence and then sort out which points that belong to which vehicle has been developed. These points are then used to estimate camera orientations and to build models of the vehicles. (Less)
Please use this url to cite or link to this publication:
author
supervisor
opponent
  • Larsen, Rasmus, DTU Informatics, Denmark
organization
publishing date
type
Thesis
publication status
published
subject
pages
145 pages
defense location
Lecture hall MH:B, Centre for Mathematical Sciences, Sölvegatan 18, Lund University, Faculty of Engineering
defense date
2016-04-08 10:15
ISSN
1404-0034
ISBN
978-91-7623-702-1
language
English
LU publication?
yes
id
b2d8cc2a-a688-4b99-891d-3316ba11620f (old id 8838253)
date added to LUP
2016-03-21 10:28:04
date last changed
2016-09-19 08:45:00
@phdthesis{b2d8cc2a-a688-4b99-891d-3316ba11620f,
  abstract     = {During the last decades, image analysis has become an important tool in various applications. The increase in computer power has led to that more advances algorithms can be developed. In this thesis image analysis has been used in three different applications; quality control of asphalt, cancer detection in histopathological images and traffic surveillance.<br/><br>
<br/><br>
In the first part of the thesis image analysis has been used to assist researchers at asphalt laboratories. Asphalt consists of a mixture of stones of different sizes and a binder called bitumen. Mainly two problems have been studied; the affinity between stones and bitumen and grain size distribution in asphalt samples. The affinity between bitumen and stone is very important for the durability of the pavement and is traditionally measured by the rolling bottle test. A goal in this thesis is to replace the manual evaluation in the method by automatic image analysis to make this test more accurate and objective. Another quality control is to estimate the size distribution of stones in asphalt samples to see if it follows the recipe for the asphalt. To avoid toxic substances, that today are used to dissolve the sample to retrieve the stones, we use image analysis to analyze cross sections of the samples.<br/><br>
<br/><br>
In the second part, histopathological images have been studied. Features from pre-trained deep neural networks have been used to classify images of prostatic tissue. These features were used to train classifiers to classify the image into four different classed; benign tissue and three different grades of cancerous tissue. Doing this automatically will help the pathologist to speed up their diagnostics in the future.<br/><br>
<br/><br>
The third and last part of the thesis deals with tracking and 3D-reconstruction of vehicles. By doing 3D-reconstructions of vehicles it is possible to accurate estimate their positions. A system to track interesting points between images in a movie sequence and then sort out which points that belong to which vehicle has been developed. These points are then used to estimate camera orientations and to build models of the vehicles.},
  author       = {Källén, Hanna},
  isbn         = {978-91-7623-702-1},
  issn         = {1404-0034},
  language     = {eng},
  pages        = {145},
  school       = {Lund University},
  title        = {Applications of Machine Vision — Quality Control, Cancer Detection and Traffic Surveillance},
  year         = {2016},
}