Task sensitive attention tracking

点击翻译按钮获取中文摘要

The goal of the research is to develop techniques for determining user task and using this information in tracking and/or predicting the allocation of attention resources. There are many existing methods for tracking attention to media such as television, web-pages or magazines, generally based on tracking of overt attention, that is, the tracking of eye movements. These methods are appropriate when the subject is carrying out a task that requires gaze to be centred on the objects of interest, such as in reading. However, these methods are of little use when the person is engaged in tasks requiring peripheral vigilance (such as in many sports or playing of video games). Instead, what is needed is covert attention tracking, or knowing where people are paying attention away from their central vision (i.e. "looking out of the corner of their eyes"). Covert attention tracking is typically done using a model of image salience, which is the specification of which visual features are most attractive to attention. Although image salience models have been extensively researched and are quite well-developed, they can, at best, only account for about 75% of the actual attention allocations by humans. It is our view that this shortcoming is due to a lack of task-dependence in the models. Thus, we propose to develop machine-learning based techniques for determining the task being carried out by a subject, in order to tune task-dependent attention models.