Human Behavior Intention Estimation of Goal Instruction Based on Saliency Map in Gaze Controlled Teaching System for Omni-Directional Wheelchair

Abstract

In recent years, many support systems intended for elderly and disabled people have been developed. Among them, the study on the interface utilizing human gaze has been attracting attention. Human vision contains a lot of important information. Some studies that estimate a next action of operator intended to perform by using the visual information have been reported. As an example, the visual saliency map model is one of the means using to estimate the intention of the human gaze. This model enables to estimate the place directed human attention from the image. Also, this model has an advantage that uses only the image information. As the research dealing with the visual saliency map model, the study suppressing the changing value of saliency map by the movement of operator’s head and the study controlling the weight of saliency map used human knowledge have been reported. In this research, we focus on the combination of visual saliency map and gaze information, and extract a behavioral intention estimation map from the landscape and the movement of human eyes. A gaze instruction system using panoramic expansion image for controlling the locomotion of omni-directional wheelchair is proposed in this paper. In this system, the human intention by various input information is estimated by using fuzzy reasoning. To evaluate the effectiveness of this method, we carried out the comparison experiments for gaze instruction system by using only movement of human eyes.