Measuring Behavior 2005: Hirvonen et al.

Comparing two fixation algorithms on eye movement data from a multitask experiment

K. Hirvonen, K. Lukander and J. Virkkala

Section of Clinical Neurosciences, Finnish Institute of Occupational Health, Helsinki, Finland

The purpose of this study was to investigate the effect of applying different fixation algorithms to the same raw eye movement data, measured while the subjects (n=3) performed a five minute multitask with three different workloads. The multitask is used to simulate typical work life situations, where employees have to divide their attention between several different cognitively demanding functional tasks. Our in-house developed Brain@Work -multitask software contains four machine paced tasks: arithmetic, memory, visual, and auditory vigilance task. Eye movements were recorded with EyeLink and the raw data from the left eye has been analyzed with EyeLink and iView software.

The EyeLink analysis algorithm uses velocity thresholding. Raw data points are identified as a saccade if either the eye velocity or acceleration values are above a certain threshold value (manufacturer’s default threshold values were used: acceleration 9500°/s2, velocity 30°/s). The iView analysis algorithm uses a spatial window. This algorithm identifies consecutive raw data points as a fixation if they are within the de.ned spatial threshold area. The defined fixation is rejected, if the fixation time is shorter than the minimum fixation time threshold value. For the .nal comparative analysis, threshold parameters of 20 pixels and 80 ms were used as these gave the best correlation with the Eyelink data.

The different analysis tools’ result data differ by the number of fixations and the fixation times. The resulting spatial data is quite similar. However, different conclusions on cognitive performance and strategies can be made depending on which of the analysis method is used. There are less and longer fixations in the EyeLink analysis results than in iView results. The iView analysis algorithm cuts long .xations to smaller fixations, because of the maximum fixation area threshold value.

Each fixation analysis tool has a characteristic way to accentuate specific features from the raw data, therefore it is essential to pick correct analysis tool for any single study.

Paper presented at Measuring Behavior 2005 , 5^th International Conference on Methods and Techniques in Behavioral Research, 30 August - 2 September 2005, Wageningen, The Netherlands.