User Identification of Touch Screen Input Using RGB-D Cameras

Abstract: An attempt was made to identify the users of a 78" curved touch screen based on images from three cameras placed around it. By assuming the existence of a robust facial detection algorithm, the work is limited to connecting each touch point to the most likely face. The cameras provide both color and depth information, allowing 3D logic to "see" that a touch point is connected to one of the detected faces. Multiple algorithms were implemented and evaluated, based on segmentation and path-finding. While the general approach is likely feasible, none of the tested solutions provided the accuracy necessary for a robust system. In order to achieve better robustness, the most promising improvement would be a better hardware configuration with more cameras.