Semantic evaluation of region of interest for intelligent robot

This paper introduces the concept of semantic evaluation of Region of Interest (ROI) for intelligent robots. The intelligent robot must have the capability of understanding situations. The first step of understanding of the situation is to find where to focus on and how to behave. Focusing on some particular area or region needs selection of the objects of interaction relevant to the context. Moreover, the focused area needs to be semantically evaluated to quantify the semantic relations. In this paper, we first detect interacting objects based on dynamic interaction. Then we recognize probable objects using Dynamic Bayesian Networks. Using the probable objects and a mutual supplementation model, we determine the contextual object. We form ROIs based on possible combinations of objects and the contextual object. Finally, we semantically evaluate each ROI. Various experimental results are provided to illustrate our method.