Premium
On‐Line Classification and Prediction of Eye Movements by Multiple‐Model Kalman Filtering
Author(s) -
Kohlbecher Stefan,
Schneider Erich
Publication year - 2009
Publication title -
annals of the new york academy of sciences
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.712
H-Index - 248
eISSN - 1749-6632
pISSN - 0077-8923
DOI - 10.1111/j.1749-6632.2009.03776.x
Subject(s) - kalman filter , computer science , computer vision , gaze , artificial intelligence , eye movement , smoothing , eye tracking , extended kalman filter
An extensible multiple‐model Kalman filter framework for eye tracking and video‐oculography (VOG) applications is proposed. The Kalman filter predicts future states of a system on the basis of a mathematical model and previous measurements. The predicted values are then compared against the current measurements. In a correcting step, the predicted state is enhanced by the measurements. In this work, the Kalman filter is used for smoothing the VOG data, for on‐line classification of eye movements, as well as for predictive real‐time control of a gaze‐driven head‐mounted camera (EyeSeeCam). With multiple models running in parallel, it was possible to distinguish between fixations, slow‐phase eye movements, and saccades. Under the assumption that each class of eye movement follows a distinct model, one can decide which types of eye movement occurred by evaluating the probability for each model.