Simulating Visual Attention of Human Driver in Different Task Engagement Modes by Combination of Moving Object Attractiveness Map and Saliency Map

Abstract

Moving object attractiveness map is a computational model that evaluates the possibility that a moving object in the field of view is gazed by the observer, which generates a view image in which each moving object is represented by shading according to the calculated gaze probability. Saliency map is, on the other hand, a computational model of visual attention that applies scientific knowledge of information processing in low-order visual cortexes to image processing, which generates an image that represents the saliency distribution based on low-level image features such as brightness and hue. This paper presents a method that combines outputs of the moving object attractiveness map and the saliency map to estimate the extent to which each area of the view image is likely to attract the visual attention of a human driver. The proposed model demonstrates that it can simulate a driver in different task engagement modes, such as watching out for surrounding moving objects or looking for a particular environmental object, by changing the weight of each of the two maps for linear combination.