Abstract
Research in our lab has been concerned with the statistics of saccade lengths and fixation durations during free viewing of natural images. However, such experiments consist of 2-D displays with the head fixed. It is not obvious that findings in this domain should generalize to natural scenarios. Bahill, Adler and Stark (1975) examined the statistics of saccade length while participants walked outdoors wearing a mobile eye tracker, but there has been no subsequent effort to replicate these findings with modern eye trackers, with more statistical analyses, and in a wider range of natural tasks. To address this, we used a mobile eye tracker to capture eye movements of participants while in engaged in a set of common tasks including: Making a sandwich, playing Frisbee, transcribing a piece of text into a word processor, navigating an office hallway, and navigating a city street. Eye-in-head position data were median filtered and segmented by a 35째/s velocity threshold and separated into saccades and fixations. We analyzed saccade length, duration and orientation, and fixation duration. We present preliminary data from a set of normally sighted subjects. Over the course of ~25 minutes total, subjects made an average of ~3000 fixations. Saccade lengths peaked at ~3-6째 degrees with a heavy tail typically ranging out to about 40째, but also including some larger saccades. We replicate Bahill et al and find that ~85% of saccades are under 15째. Additionally, while saccades occur in all directions, the orientation of saccade trajectories is generally biased towards the cardinal directions, occurring roughly twice as often as other orientations. Fixation durations had a median of ~400ms with a long tail with extending out to ~4-5s. This data set provides a reference for normative human eye movement behavior in non-constrained circumstances.
Meeting abstract presented at VSS 2014