Automating the Segmentation of Eye Movements in a Robotic Virtual Reality Environment

Modern eye-tracking technology can record eye movements at a high temporal and spatial resolution during behavioral tasks, allowing researchers to make inferences about sensorimotor integration and visual processing. Analysis of the gaze data often requires segmentation into fixations, saccades, and smooth pursuits; a process that is very labor intensive when annotated manually. Segmentation methods have been developed to automatically process user experience and reading data, but most use data from subjects viewing static stimuli and are not suited to gaze data containing smooth pursuits. A novel method using hidden Markov models is proposed to automate eye movement classification in virtual reality and robotic environments used in visuomotor neuroscience research. This method performed with 72 percent accuracy when vergence eye movements were considered and 82 percent accuracy when they were excluded.