Exogenous and Endogenous Based Spatial Attention Analysis for Human Implicit Intention Understanding

摘要

In this paper, we develop a novel human implicit intention understanding model by mimicking the human-like visual attention and brain information processing mechanisms. In other words, the proposed model considers a hybrid cognitive neural system, which comprises of spatial attention model obtained based on exogenous and endogenous attention models. Generally, information can be selected via top-down or endogenous mechanisms depending on the goals of the observers while salient objects or events attract spatial attention via bottom-up or exogenous mechanisms allowing a rapid and efficient reaction to unexpected but important events. Given a visual stimulus, the spatial analysis module identifies the objects of interest by correlating the salient areas obtained from the exogenous module and the eye gaze information obtained from the endogenous module. Then, corresponding to an intent, each of the identified objects are classified in to one of the two classes - intent object or non-intent object, by analyzing the features such as fixation length (FL), fixation count (FC) and pupil size (PS) corresponding to each object. In the proposed model, support vector machine (SVM) is trained for classifying the different objects. Experimental results show that the proposed model generates plausible performance based on hybrid cognitive neural system.