Attention determination for social robots using salient region detection

Abstract

It is amazing that human beings can attend to the interest from a simple glimpse of scenes. Similarly, it is also important for social robots to determine its attention autonomously so as to behave human-likely and naturally. In this paper, we propose a technique to model this biological ability by searching saliency regions in a fast and reliable way. The salient regions are detected based on information entropy and biological color sensitivity. The information entropy evaluates the level of knowledge and energy contained, and the color sensitivity measures the biological stimulation to eyes of the presented scene. The performance of the detector is studied on natural scenes. The experiments proved the effectiveness of the detector and the important properties of invariance to transformation and illumination. © 2010 Springer-Verlag.