Ballard D H, Hayhoe M M, Salgian G, Shinoda H
Department of Computer Science, University of Rochester, NY 14627-0226, USA.
Spat Vis. 2000;13(2-3):321-33. doi: 10.1163/156856800741144.
One of the most widely used terms in the study of human performance is attention. Yet it can also be argued that it is one of the most confusing and misunderstood. Huge variations in performance, from not noticing large changes in images or natural situations, to differences of tens of milliseconds have all been described as attentional effects. We argue that the large disparity in results can be more easily understood in the context of a fairly complete model of human performance that describes the execution of a set of complex natural tasks via a collection of visual routines that extract crucial information from the optical array. The description of visual routines is hierarchical. At the most abstract level, a scheduler must pick a small set of programs for the current tasks. Each program contains steps which are keyed to information in the scene. This is extracted by visual routines which run during a single fixation and extract pertinent information. The library of routines themselves are designed to execute quickly, but their actual performance depends on signal-to-noise characteristics of the imaged scene. The hierarchical description of behavior shows that questions about attention make sense in the context of the descriptive level in which they are embedded. We illustrate these principles with examples of driving behaviors.
