Introduction
Observation of animal behavior is important to many fields of research in the life sciences. Its automation is crucial to advances in animal well-being and scientific research. In the EthoVision product made by Noldus, a Windows computer is attached to a camera which has a 2D top view of a home cage. The images are recorded and analyzed in real-time to generate statistics on what the animals are doing. In many cases the lack of contrast and/or reflections in the environment cause a poor visibility of the animal. We solve this by using model based approaches but, like background modeling, but these have their shortcomings.

Assignment 
The aim of this assignment is to identify a method able to reliably find the correct contour of the animal. Segmentation algorithms known from literature can facilitate the identification of an object in an image or video. The assignment is to propose an optimal combination of preprocessing filters and segmentation techniques - based on theory and experimentation. Such techniques will likely include methods such as ‘region growing’, ‘watershed’, ‘level set’, ‘pyramid’, e.g.

Your background
We are looking for students who want to specialize in the area of computer vision and recognition. Our tools include Matlab and C++.  

More information
If you are interested in this particular project, please contact Rob Ottenhoff or Wil van Dommelen. Background information: www.noldus.com.