Abstract
We are usually watching a movie while predicting the next scene. In this paper, we observe this phenomenon in detail and propose a new model of these characteristics. After that we prove this hypothesis is correct by experiment. The aim of our model is to clarify the characteristics of human prediction and to estimate human gaze motion automatically by detecting the object which is the most important in current scene, the object which will be the most important in the next scene, and the event information between them.