Abstract
The validity of eye-movement-based research rests on the assumption that the algorithms applied to variously sampled signal correctly characterise small changes in eye behaviour. Despite this fundamental assumption, a systematic comparison of algorithms and filters across typical data quality, as measured from various commercial systems, is lacking. Comparing systems and algorithms for accuracy of event characteristics was previously approached through a) dual-recording of the same eye with two systems, e.g. VOG and coil, which is limited due to the different nature of each signal type b) recording with 2xVOG systems concurrently, which presents problems with system IR illumination profiles, c) recording each system separately in a controlled test environment with a robotic artificial eye working comparably on all systems, reaching the velocity of real eye movements, or d) events based on expert hand-coding of recorded data as 'gold standard' – requiring that experts agree. This study takes the novel approach of recording optimal data on a DPI eyetracker during fixation, saccade and smooth pursuit of varied amplitude and velocity, recording the same participants on several VOG systems to model spatial and temporal noise, and applying the error measured from each participant to comparative DPI datasets. We compare the resulting datasets to the original DPI data for fixation number, fixation duration, saccade amplitude, velocity, acceleration and peak velocity characteristics following the application of several current filter and event detection algorithms. The results provide a guide to improved eye movement measures from a particular study, and improved system performance, based on objective analysis. Results are discussed in terms of their relevance for anyone doing research with eye trackers, as well as for those developing systems and analysis methods. It contributes towards a larger study with the aim to bring clarity regarding which measures are valid from various types of eye trackers.
Meeting abstract presented at VSS 2014