Identifying Human-Object Interactions in Motionless Images by Modeling the Mutual Context of Objects and Human Poses

The two most challenging problem in computer vision is to detect an object in clutter view and estimate articulated human body from 2D images. While the difficulty occurs during the activities which involves human objects interaction, since the related items tend to be very small or moderately visible and the human body parts are often self occluded. We examine that the human pose and the object are share joint background to each other, i. e. if we recognize one, make easy to recognize the other. In this paper, we proposed the mutual context model for jointly modeling object and poses of human body in the human object interaction activity. In this method, recognition of object provide the strong prior for better human pose estimation, this human pose estimation enhance the efficiency of detecting the objects that interrelate with the human.