Eye tracking is a useful tool for analyzing what a person is interested in, detecting which context their are in, and for novel interactions. In order to make sense of the eye data, eye movement detection algorithms are necessary. The main classes of eye movements are fixations (when our eyes are still on an object of interest), saccades (the quick jumps our eyes makes when going from a fixation to another) and pursuits (the smooth movement our eyes perform when following a moving object). There has been a great number of algorithms that allow the detection of fixations and saccades, but when I started my Ph.D there was no way to reliably detect smooth pursuits.

I developed an algorithm for this purpose, following a machine learning approach. I collected a large number of eye movements to form a database. I then developed a new set of features to extract from the signal. These are called shape features, because their value reflects the global shape of the signal, in order to detect the specific pursuit pattern. Shape features are based on the evolution over time of traditional features such as speed, range, etc.

The algorithm is done, and I am now working on its evaluation to show that it is versatile and capable of recognising pursuits in real time and in different settings.



Papers:

Detection of Smooth Pursuits Using Eye Movement Shape Features,
M. Vidal, A. Bulling and H. Gellersen, Proc. of the 2012 Symposium on Eye-Tracking Research & Applications (ETRA ’12). March 2012.
PDF | DOI
Analysing EOG Signal Features for the Discrimination of Eye Movements with Wearable Devices,
M. Vidal, A. Bulling and H. Gellersen, Proc. of the 1st International Workshop on Pervasive Eye Tracking and Mobile Eye-Based Interaction (PETMEI 2011). September 2011.
PDF | DOI