Thesis Project Form

Marker-less motion analysis on 2D RGB video data is becoming a reality in the computer vision community. Indeed, deep learning architectures are producing reasonably accurate estimates of body joints locations, which can be used to derive a kinematic model of a body or a body part, and possibly identify anomalies. Furthermore, optical flow is a very powerful technique that allows the extraction of parameters that describe the motion in videos. In this thesis we will apply these techniques in order to study human motion analysis in a specific scenario: videos of people with epilepsy and parasomnia.

The long-term goal of this project is the characterization of quantitative parameters that should allow the description of human motion patterns of epilepsy events. In particular, we want to analyze videos of people with epilepsy and parasomnia in order to distinguish between the two cases. In order to accomplish this goal, the proposed thesis has different aims:
- the research of parameters that could quantitatively describe epilepsy event from videos;
- the extraction of the meaningful parameters that could help the distinction between epileptic people and parasomniac ones;
- the supervised and unsupervised classification of the two different groups.

The student will learn:
1. Computer vision techniques in order to analyze images and videos;
2. How to use algorithms based on deep learning to estimate the pose of the people in the images;
3. To correlates data;
4. Machine Learning techniques that will allow the clustering and the classification of the data;
5. Improve the knowledge of Matlab.