Master's Thesis Projects
The group supervises about 5-15 master's thesis projects per year. Many of these are done directly at the department, close to the research front. Several of the theses at the department have lead to high-quality peer-reviewed publications. Also, some of the theses are done in collaboration with industry. If you are interested take a look at this designated home page. Feel free to come and discuss your own ideas for master's thesis projects with Carl Olsson, Kalle Åström, Fredrik Kahl, or Magnus Oskarsson.
Seminars
Project Proposals
Note that some proposals are written in Swedish and some in English.
Automatisk inspelning av sport
Många moderna digitala kameror kan styras automatiskt (PTZ, pan, tilt och zoom) och ofta kan relativt avancerad mjukvara köras direkt i kameran. Detta projekt syftar till att utveckla mjukvara som automatiskt kan filma, följa och mäta tider för ett idrottsevenemang, till exempel ett 100m-lopp i friidrott. För detta krävs dels kunskaper i matematisk bildanalys, men en del kunskaper i programmering kommer även att behövas. Vi får hjälp av IFK Lund att spela in film och utvärdera resultatet. Inom detta projekt finns möjligheter för flera examensarbeten.
- Ett examensarbete syftar till att utveckla algoritmer för automatisk följning av lopp. Från början är kameran stilla och efter starten skall den hela tiden följa löparna genom att rotera och zooma. Algortimen måste kunna kompensera för eventuella rörelser i publiken och den egna kamerarörelsen.
- Ytterligare ett examensarbete behandlar automatisk detektion av löparbanan. Genom kamerakalibrering och banans kända mått kan löparnas hastighet beräknas och diagram över hur hastigheten förändrats genom loppet presenteras för publik och tränare. Denna information efterfrågas av aktiva löpare. Automatisk detektion av mållinje är även viktigt för tidtagningen ovan.
Kontaktpersoner:
Petter Strandmark
Håkan Ardö
Kalle Åström
Maximum Flow/Minium Cut for Grid Graphs
The classic problem of finding the minimum cut, or equivalently, the maximum flow in a graph is a problem with many applications in computer vision and image analysis. The graphs are often regular and there exists very efficient algorithms for computing the minimum cut for these graphs.
However, the implementations commonly used are general and use more memory than would be necessary if the regular structure were exploited. The goal of this Master's thesis is to implement a memory-efficient version, significantly reducing the memory requirements, while retaining as much speed as possible.
The student needs to be familiar with C++.
Contact: Petter Strandmark or Johannes Ulén.
Recognition of Everyday Objects
Consider a question like "Where is my
bicycle?" or "Where are my keys?". A person's bicycle is a personal everyday object, which will
be given a place in the personal memory. The aim of this project is to investigate and implement methods for answering such questions given images or video sequences where these kinds of objects are present. More specifically this entails both detecting and recognizing objects in images. This project will be part of a system that we are developing within a research project in our group, called Wearable Visual Information Systems. The ease with which pictures are captured and stored today opens up many new possibilities for exploiting captured visual information for the purposes of content analysis, guiding, information access and memory enhancement. It is the purpose of this project to advance these tools and show their usefulness in a set of compelling demonstrators based on personal wearable image capturing devices. See also:
Wearable Visual Information Systems.
Contact: Magnus Oskarsson.
Optimization Techniques for Computer Vision and Image Analysis
One of the research areas in computer vision and image analysis that has
received a lot of attention lately concerns efficient optimization techniques. We have developed several state-of-the-art methods for multiple view geometry, correspondence problems, image segmentation and inpainting, and we are currently working on generalizing and improving these techniques for more settings.
Contact: Fredrik Kahl or Carl Olsson.
SLAM on a quadrotor helicopter
Simultaneous localization and mapping (SLAM) is a technique used by robots and autonomous vehicles to build up a map within an unknown environment (without a priori knowledge) while at the same time keeping track of their current location.
The Mathematical Imaging Group owns a quadrotor helicopter (AR Drone) with a built-in camera which is possible to control from a computer while recieving video. The goal of this Master's thesis is to develop a SLAM system for this helicopter. Interesting problems to solve include estimating uncertainty and fly to a good location in order to reduce it.
Apart from having participated in the courses in Image Analysis and Computer Vision, knowledge of a suitable programming language required to write software to control the helicopter. The Imaging Group has a working interface written in Python, but interfaces in C, C# and Java are available online.
Contact: Petter Strandmark or Sebastian Haner.
myTown
The aim of this project is to model and visualize (part of) a city from an unsorted photo collection, where the collection is built up collaboratively by users of a web community myTown. By making use of modern feature matching teqniques and knowledge of camera 3D geometry, it is possible to find correspondences between images in an unsorted collection of photos taken in the same geographical area. With this information one can build a "camera graph" where edges correspond to cameras with overlapping field of view. Using this graph structure it is then possible to visualize and navigate the photo collection in 3D by jumping from image to image. The basic idea is very similar to Microsoft's Photo tourism/photosynth with some key differences:
- The images will be contributed incrementally by any user who is part of myTown via a web interface.
- The images will probably cover a much larger area than normal photosynths and with a much sparser structure.
- There is likely to be additional meta information such as gps position or cell id if the images are taken with a mobile camera.
- The total number of images is potentially huge which is likely to make both image to image matching and geometry computations very challenging.
Suitable master thesis topics will probably consist of parts of this project. There are good possibilities for both theoretical and more practically geared theses including web design and visualization.
Contact: Martin Byröd.
Visuell odometri och navigation för GPS-störda miljöer
I samarbete med Security and Defence Solutions, Saab, finns ett antal förslag på examensarbete inom bildbaserade metoder för navigering och odometri, speciellt i miljöer där GPS inte fungererar.
Placering: Huskvarna.
Contact: Fredrik Kahl.
Object Recognition of 3D Models
This is a classical problem in computer vision: How to recognize an object in an image? In our research group, we are currently developing several new and efficient methods based on alignment and the aim of this thesis is to explore and investigate such techniques for recognizing rigid 3D objects.
Contact: Olof Enqvist or Fredrik Kahl.
Medical Imaging
There are several possible topics within medical imaging, for example: 1. Segmentation of MR images. In our research group, we have developed new optimization methods for cardic MR segmentation and we are interested in generalizing these techniques further for new medical applications. 2. Multiple color stainings of cells. A new type of method for detecting and diagnosing cancer in lung tissues has been developed at the unit of Airway Inflammation and Immunology, Lund University Hospital. There are several unsolved image analysis problems for increased automation of such cell analyses. 3. For detecting prostate cancer, a method for reliable registration of neighbouring sections from a prostate biopsy sample is needed. Due to different sensor modalities of the imaged samples, it is difficult to automatically register the different sections of the sample. Collaboration with Department of Urology, Malmö University Hospital.
Contact: Fredrik Kahl, Niels Christian Overgaard or Olof Enqvist.
Completed Projects
- Automatic Segmentation of Menisci in MR Images Using Pattern Recognition and Graph Cuts
- Semantic Scene Classification for Enhanced Image Browsing Experience
- Identifying and connecting to screens using mobile phone cameras
- A System for Automated Hand Pose Estimation in Neurophysiological Research
- Automatic Registration of Large Images from Light Microscopy
- Human Pose Estimation using Dynamic Programming
- Distributed Mobile Computer Vision and Applications on the Android Platform
- Trajectory Analysis in Surveillance
- Finding representative structures in large image collections
- A System for Real Time Gesture Recognition
- Automatic Detection and Segmentation in Lymphoma with PET/CT
- 3D Reconstruction for Traffic Surveillance
- Feature Tracking with Multiple Models
- Evaluation of methods and configuration for constructing a people detection system
Resources
Questions: webmaster
Last updated: 2012-01-24
- Conference paper:
Curvature-Based Regularization for Surface Approximation
Carl Olsson
Yuri Boykov - Conference paper:
An Automated System for Analysis of Renal Scintigraphy Images
Matilda Landgren
Karl Sjöstrand
Lars Edenbrandt - Conference paper:
Covariance Propagation and Next Best View Planning for 3D Reconstruction
Sebastian Haner
Anders Heyden