Queueing network modeling of visual search

Queueing network mental architecture represents the human cognitive process in the form of a queueing network with multiple servers, entities, and routes. Servers represent functional units in the human brain; entities represent pieces of information to be processed by the servers; and routes represent flows of information between the servers in the system. The Queueing Network - Model Human Processor (QN-MHP) is a computational model of the queueing network mental architecture, and the basis of the computational cognitive models developed in this dissertation.; Two problems were studied in this dissertation: (1) the nature of the underlying mechanism to develop eye movement strategies in visual search and (2) the minimization of task-dependent properties in the model. A series of computational cognitive models were developed and examined by comparing their performances with the human experimental data of random menu search, picture viewing, color-shape conjunction visual search, and pedestrian detection task.; This dissertation consists of five studies. In the first study, a visual stimulus conversion system which transfers a digitized natural scene to data arrays representing attributes of a visual stimulus in the scene was developed. The second study presents the QN-MHP model of random menu search. The third study introduces the reinforcement learning process to explain human eye movements. The fourth and fifth studies explore the effects of a concurrent task on visual search. The experimental results of pedestrian detection using night vision enhancement systems while driving were analyzed in the fourth study. In the fifth study, the dual tasks of pedestrian detection and driving were modeled based on the QN-MHP.; To obtain eye movement strategies for visual search, the reinforcement learning process was integrated into the QN-MHP model. To minimize task-dependent properties in the model, the number of task-specific production rules to reproduce human-like behaviors was minimized by using a structural characteristic of the queueing network architecture. Further, simulation of the pedestrian detection task in driving, demonstrated that the QN-MHP can be used to study complex visual tasks in a real world setting as well as simplified controlled laboratory tasks.