Event Detection Algorithms and Data Quality

Description

Eye Tracking research is heavily dependent on event detection algorithms. There are numerous algorithms, using different approaches, and they partly produce very diverse results. While it is well documented that different algorithms can produce dramatically different results (see, for example, Nystrum and Holmqvist, 2011), often, little attention is paid to algorithms in research, with most studies conducted using the proprietary algorithm packaged with the eye tracker. This makes eye tracking research difficult to compare and reliably replicate.

This talk will present a comparison of various event detection algorithms used in regard to still scene viewing. We used data from a study about the cultural influences on art perception, performed by the Lab for Cognitive Research in Art History at the University of Vienna and the Aesthetic Science Laboratory at the Keio University (Tokyo). The aim of the study was the comparison of an Austrian and a Japanese group. However, it turned out that the results of the comparison are highly dependent on the event detection algorithm. We therefore compared one algorithm by Marcus Nystrum and Kenneth Holmqvist (Nyström and Holmqvist 2010) and two by researchers at the Perception Engineering Group at the University of Tubingen (Tafaj et al., 2012) based on our own research and previous studies done on algorithm comparison (Anderson et al, 2017). The talk will address methods of comparison, qualities of each algorithm, and the drastic differences in data they produced.

Period	20 Aug 2019
Event title	20th European Conference on Eye Movements
Event type	Conference
Location	Alicante, SpainShow on map
Degree of Recognition	International