Design and implementation of a head-pose estimation system used with large-scale screens

In this paper, we propose a novel head-pose estimation system for use with a large-scale screen to provide intelligent interaction with content. The head image of the user is captured from a RGB-D (red, green, blue pixel value, and depth data) camera connected to a large-scale display system. The head orientation of the user is then estimated from the RGB-D data by using the random regression forest algorithm. The random regression forest algorithm is a very powerful tool for generalization problems that does not suffer from overfitting. By using the head-pose estimation system, the user's region-of-interest (ROI) is found in a large-scale screen. After the ROI is found, various intelligent interactions with content can be possible. As future work, a hand gesture recognition system will be jointly connected with this head-pose estimation system in order to control the user's gestures more precisely in the ROI.