- Realtime scene analysis in infrared images
Student: David Forslund
Supervisors: Per Cornvall and Jacob Roll
Subject supervisor: Joakim Lindblad
Publisher: CBA Master Thesis No. 106 / UPTEC F09 010
Abstract: The vehicle industry is, as society in general, evolving towards increasing machine intelligence. A large step in vehicle intelligence would be for the vehicle to be aware of the scene in which it is located. Scene classification is a growing field in image analysis, and much progress has been made in the last few years. This thesis aims at employing scene classification in real-time to images constructed from an infrared sensor that is mounted at the front of a vehicle. The images are greyscale, and the specific task studied is the two-class task of separating city and rural road scenes. Several image representation methods for scene classification, such as 'Edge Direction Histograms', and 'Invariant Moments' have been evaluated, but most focus has been turned toward the 'Bag of Words' algorithm for image representation. This algorithm has been implemented using both sparse and dense extraction of image elements and using both descriptors based on simple representation of square grey patches and the popular SIFT representation. Using the 'Bag of Words' algorithm, a method for fast two-class scene classification, suitable for a real-time application is proposed. The proposed method uses grey patch based image elements, a modified floating search algorithm for visual word selection and an SVM classifier for the final classification.
- A Sense-and-Avoid Algorithm for Manned Aircrafts
Student: Erik Ahlberg
Supervisor: Harald Klomp
Subject supervisor: Gunilla Borgefors
Publisher: CBA Master Thesis No. 107 / UPTEC F09 021
Partner: IMINT Image Intelligence, Uppsala
Abstract: As the civilian air traffic keeps on growing, new and innovative solutions are needed
to handle the increase of traffic, especially with regard to safety. This is emphasized as
unmanned aerial vehicles are currently in the verge of being integrated into the
already highly congested general airspace. An autonomous sense-and-avoid system has
the possibility of revolutionizing air safety by offering additional safety measures for
manned aircrafts and allowing for unmanned aerial vehicle integration. In this paper
we explore this technology further and commit to the design of a sense-and-avoid
system centred on the electro-optical hardware technology. The aim is to achieve
successful detection and tracking of objects from an image sequence. This is achieved
by object detection from motion and tracking through the implementation of a linear
Kalman filter.
By extensively using an open-source public computer vision library and off-the-shelf
hardware components we managed to keep the solution relatively inexpensive.
Simulations on two separate scenarios were performed with mixed results. Although
the performance of the constructed algorithm did not completely meet the
expectations laid out before it, the simulation results showed great promise for future
improvements.
- Automated Methods for Generation of Input Function in PET Studies Using MVW-PC Images
Student: Johan Olsson
Supervisor: Pasha Razifar
Subject supervisor: Ewert Bengtsson
Examiner: Anders Jansson
Publisher: CBA Master Thesis No. 108 / IT 09 001
Partner: GE Healthcare, Uppsala
Abstract: Modeling is an approach for extracting quantitative values from PET. The signal from a
reference region or from blood samples is used as reference. Since blood sampling is
risky, this report presents an automated method based on MVW-PCA for using blood
data from the images.
The study was performed on clinical PET data from several human brains using the
tracer PIB. Two veins were found in a MVW-PC and an average of the TACs from
the relevant locations was formed. Finally, a correcting function was calculated.
The curves generated from the image data were very similar to the curves generated
from blood samples, with the largest errors in the beginning of the scan.
The used method shows potential for generating very good results if worked on
more. One of the strengths of the approach is that it is not limited to a specific tracer
or time protocol, since the MVW-PC will be chosen depending on the weights for the
first 60 seconds.
- Low Cost Real-Time Gaze Tracker Using a Web Camera
Student: Christer Bergman
Supervisor: Gustav Öqvist
Subject supervisor: Ewert Bengtsson
Publisher: CBA Master Thesis No. 109 / UPTEC F09 048
Partner: Bernadotte Laboratory, St. Erik Eye Hospital
Abstract: Gaze tracking means to detect and track the direction in which a person look. Gaze tracking can for instance be used in human computer interaction and in medicine. Most gaze trackers today are expensive and some are invasive. The idea of this thesis is to make a cheap and user friendly gaze tracker.
The approach is to use feature based image analysis to find and track the eyes, Hough circle transform is used to find the eyes. A simple calibration is done to get positions on the screen as output.
An experiment is done on ten volunteers to evaluate the gaze tracker. The experiment shows that the gaze tracker is robust and its resolutions are sufficient for some applications.
- k-Uniform Tilings by Regular Polygons
Student: Nils Lenngren
Supervisor: Gunilla Borgefors
Subject Supervisor: Vera Koponen
Publisher: CBA Master Thesis No. 110 / U.U.D.M Project Report 2009:23
Abstract: k-uniform tilings by regular polygons are tilings with k
equivalence classes of vertices with respect to the symmetries of
the tiling. The enumeration of all k-uniform tilings for
specific values of k is a far from trivial problem. In this
paper, I review the most important steps in the investigation of
k-uniform tilings: Kepler's enumeration of the 1-uniform
tilings in 1619, the rediscovery of the 1-uniform tilings by
Sommerville in 1905, the complete enumeration of the k-homogeneous tilings (a set of tilings which includes the
1-uniform and 2-uniform tilings) by Krötenheerdt in 1969-1970,
Chavey's enumeration of the 3-uniform tilings in 1984, and
Galebach's enumeration of the 4-uniform, 5-uniform and 6-uniform
tilings in 2002-2003. I also discuss some approaches that might be
used in future work on k-uniform tilings.
- Signal Extraction and Separation in PET Studies With Different Time Protocols Using MVW-PCA
Student: Fredrik Engbrant
Supervisor: Pasha Razifar
Subject Supervisor: Ewert Bengtsson
Examiner: Tomas Nyberg
Publisher: CBA Master Thesis 111/ UPTEC F09018
Partner: GE Healthcare, Uppsala
Abstract: Positron Emission Tomography (PET) is a non-invasive imaging modality used to
visualize the functionality in tissues and organs in vivo in medical and research
applications. PET is based on measuring the concentration of a tracer molecule,
labeled with a radionuclide, designed to follow a specific physiological and
biochemical path. PET can be used to detect neurological disorders such as
Parkinson's disease, Alzheimer's Disease, phobia and schizophrenia by studying
dynamic PET image sequences. However the PET data suffers from noise and the
different areas and tissues can be hard to discern. Previously it has been shown that
the application of Masked Volume Wise Principal Component Analysis (MVW-PCA)
can be used to separate and extract the different kinetics of the PET tracer. The
time protocol in a regular dynamic PET scan cannot be changed after the scan. The
list mode storing option, available on some newer PET cameras gives the opportunity
to reconstruct the data into any or several different time protocols after the scan. It
is important for any method used to study PET data to be more dependent on the
signal than the choice of time protocol. In this study the result from MVW-PCA using
different time protocols have been studied. The result showed that MWV-PCA is a
stable method, depending more on the acquired data than on the frame lengths.
There is however a possibility to improve the result slightly by using shorter frames,
yielding the result that similar kinetics are more clearly separated.
- Characterization and Reduction of Noise in PET Data Using MVW-PCA
Student: Per-Edvin Svensson
Supervisor: Pasha Razifar
Subject Supervisor: Ewert Bengtsson
Examiner: Anders Jansson
Publisher: CBA Master Thesis 112/ IT 09 003
Partner: GE Healthcare, Uppsala
Abstract: Masked Volume-Wise Principal Component Analysis (MVW-PCA) is used in Positron Emission Tomography (PET) to distinguish structures with different kinetic behaviours of an administered tracer. In the article where MVW-PCA was introduced, a noise pre-normalization was suggested due to temporal and spatial variations of the noise between slices. However, the noise pre-normalization proposed in that article was only applicable on datasets reconstructed using the analytical method Filtered Back-Projection (FBP). This study aimed at developing a new noise pre-normalization that is applicable on datasets regardless of whether the dataset was reconstructed with FBP or an iterative reconstruction algorithm, such as Ordered Subset Expectation Maximization (OSEM).
A phantom study was performed to investigate the differences of expectation values and standard deviations of datasets reconstructed with FBP and OSEM. A novel noise pre-normalization method named "higher-order principal component noise pre-normalization" (HOPC noise pre-normalization) was suggested and evaluated against other pre-normalization methods on both synthetic and clinical datasets.
Results showed that MVW-PCA of data reconstructed with FBP was much more dependent on an appropriate pre-normalization than analysis of data reconstructed with OSEM. HOPC noise pre-normalization showed an overall good performance with both FBP and OSEM reconstructions, whereas the other pre-normalization methods only performed well with one of the two methods.
The HOPC noise pre-normalization has potential for improving the results from MVW-PCA on dynamic PET datasets independent of used reconstruction algorithm.
- Image Analysis for Marker-less Facial Tracking
Student: Wenlan Yang
Supervisor: Christian Sjöström
Subject Supervisor: Ewert Bengtsson
Examiner: Anders Jansson
Publisher: CBA Master Thesis 113/ IT 09 042
Partner: Imagination Studios
Abstract: Tracking of facial features is increasingly used in game and film industry as well as for
other applications. Most tracking systems are currently using markers which
unfortunately are tedious and cumbersome. Marker-less facial tracking is supposed to
eliminate the disadvantages of the marker-based approaches. This thesis investigates
different algorithms for marker-less tracking and presents how to apply them in a
robust way. View-based and component sensitive normalized face images can achieve
accurate tracking results based on the Active Appearance Algorithm. Post processing
the parameters of global motions of the model smoothes the synthesized video
sequence. Tracking results for faces and a tool developed for creating training images
are also presented.
