Computational modelling of directed attention fatigue

• Main Author: Toh, Chia Ming
• Format: Thesis
• Language: English
• Published: 2015
• Subjects: QP Physiology
• Online Access: https://eprints.ums.edu.my/id/eprint/18943/1/Computational%20modelling.pdf; https://eprints.ums.edu.my/id/eprint/18943/

Our ability to concentrate on tasks is crucial for success and survival. In varying degrees, concentration is required for most activities and, for many sustained tasks such as driving, piloting and visual surveillance, a loss of concentration may be safety critical. When task concentration is sustained, performance may vary over time in response to changes in motivation, interest level and knowledge. But, even with high motivation, there are limits on how long task concentration can be sustained. Eventually, concentration will deteriorate, with consequences ranging from discomfort and performance degradation to possible injury and death. Attention, a key aspect of human perception, is the mechanism underlying concentration. A failure of attention is a failure of concentration. In the visual modality, which this project is concerned with, attention directs gaze, and failure to attend to task-relevant locations in a scene is likely to yield poor performance. Attention is usually considered to have two main modes of operation distinguished by effort and intentionality. In bottom-up mode, our gaze is drawn involuntarily to locations by salient visual properties of the scene. In top-down or voluntary mode, we choose where to look, usually in accordance with task demands. This choice means inhibiting competing stimuli and bottom-up cues, which requires effort, a resource considered limited. The inability to inhibit bottom up cues and actively direct gaze, induced by sustained concentration, has been called Directed Attention Fatigue (OAF). DAF is likely to be a key component of performance deterioration over time but its underlying mechanisms are not well understood, nor are the gaze characteristics associated with it. Computer models may be able to help us better understand OAF and might allow us to redesign tasks or performance strategies to mitigate it. However, although computational models of visual attention have been developed, no current model fatigues. This project develops a computational model of OAF by substantially extending the influential bottom-up attention model of Itti and Koch (IKM). A functional mechanism for DAF is proposed and implemented within a version of IKM heavily extended to model sustained task performance, with the addition of foveation, top-down task relevance, object recognition, decision making and action. Human data on DAF within a sustained task (custom-designed to be spatially challenging) was gathered using an eye tracker and custom software. This human data yielded insights into the gaze characteristics of OAF and was used to test the predictions of the overall model, with encouraging results for the proposed OAF mechanism.