The attention function has been classified into (i) sustained attention, (ii) selective attention, (iii) controlled attention, (iv) distributed attention, and (v) capacity for attention. Ordinarily, in order to evaluate the function of attention, the digital cancellation test (D-CAT) or trail making test (TMT) are employed. However, these evaluations are in the form of paper tests, and cannot effectively evaluate small changes in activities daily of living in real time. Accordingly, we focus on human behavior during the act of cooking, in which multitasking is common and many types of ordinary movements are performed. We previously proposed a method of evaluating distributive attention during cooking and showed its effectiveness. In this study, a method for evaluating sustained attention is proposed. Specifically, we acquire an egocentric video by an egocentric vision and estimate line-of-sight based on the visual attention model that is integrated from the three kind of models (category-specific attention, top-down attention, and bottom-up attention). After that, we define line-to-sight information from high saliency area on the visual attention model.