Improved human interaction in telepresence robot using real-time face segmentation

This paper focuses on improving human to human interaction by removing a nonverbal modality i.e. communicator´s environment. In the telepresence robot application where the user´s image is projected to a remote space, the user´s background/environment might not be valid to the on-going conversation. The proposed system is a real-time face segmentation system that extract face region from a face image. In the telepresence robot application, the face segmented is input to the 2D display for showing the user´s face. With the background removed, the unnecessary environment information that imposes unnecessary meaning onto the interactant´s decision can be minimized. The result of the system shows good robustness towards a few properties that are useful in telepresence robot application. However there are dependency and assumption in this system which can cause disturbance to the system in certain conditions. Still in the telepresence robot context, these disturbances can either be avoided or is non-destructive. Lastly, a future work is proposed to gather quantitative data on the improvement of the new proposed telepresence system.