There are some techniques of video compression based on the foveation image coding and processing algorithms. One of the most frequently asked questions is: "How do we know the foveation points?" In this paper a computational model of automatic localization of the foveation point is proposed. Till now, except for a few cases, the research related to foveation points determination has mainly concerned with the analysis of static images. The proposed model, on the contrary, simulates the gaze shifting of a human observer when the scene in development is dynamic. Through the selective attention the most relevant part of an image is selected, the so called "focus of attention" (FOA).