Zygmunt L. Szpak
Computer Vision & Machine Learning

Multiple View Geometry

There are many applications where it is necessary to determine how one view of a scene is related to another view. Based on a particular camera model and geometric arguments it is possible to devise a mathematical model of how views of a scene are related. The task then becomes that of estimating the parameters of that mathematical model.

Click here to enter the multiple view geometry project page.

Ellipse Fitting

Like the circle, the ellipse is one of those primitive geometric shapes that crops up time and time again in many different scenarios. Whether we are tracking an object, calibrating a catadioptric camera or studying cells and crystals, we are bound to end up having to fit an ellipse sooner or later. Surprisingly, fitting an ellipse is actually a hard problem, and much can be learned about more complicated parameter estimation problems in multiple view geometry by studying ellipse fitting.

Click here to enter the ellipse fitting project page.

Maritime Surveillance

I have spent some time working on a computer vision system for maritime surveillance. Pirate attacks have recently caught the media's attention, and nowadays many ships try to avoid the coast of somalia.
The greatest challenge for a maritime surveillance system is coping with the dynamic nature of the outdoor environment, the fact that the camera is usually not stationary and very small poorly contrasted targets.

Click here to enter the maritime surveillance project page.

Machine Learning

Frequently we would like to teach a computer to recognise patterns in data. For example, we may wish to build a classifier that can recognise objects in images. To achieve this there are usually many possible models that we can apply. The question then becomes, which model to choose? Furthermore, what can we say about the quality of our model?

Medical Image Processing

Computer Vision techniques form the basis for a variety of computer software that can be used as a computer aided diagnosis tool for medical practitioners. One area of application is in opthalmology where it is important to monitor the retinal vessels to detect the onset of glaucoma or diabetic retinopathy. The first step in this direction is to devise a method of automatically segmenting the retinal vessels since this is a very time consuming task if done manually.


last update: Tue Oct 06, 2015