Abstract
There has been significant recent interest in measuring neural and cognitive responses during dynamic scenes and video. In particular, it has been noted that the attention of different observers often synchronises, and that such moments can be identified when participants' eye fixations converge in space and time. The current studies aimed to establish whether convergence in fixation locations was associated with explicit self-reports of "important" moments. In two experiments, participants' eye movements were recorded while they watched a set of movie clips featuring a range of visual content. We developed a novel method for quantifying the inter-observer convergence, based on ROC analysis, and applied this method to find the moments when participants were most likely to be looking in the same place at the same time. These moments occurred more often than expected by chance, and were highest for clips involving action. Critically, there was a reliable correlation between inter-observer agreement and explicit self-reports of important moments, indicating that attention converged at meaningful times. Additional analysis of the visual and semantic content at fixation allows a data-driven approach to video, and provides a rich source of information for those investigating natural, active vision.
Meeting abstract presented at VSS 2014