Human computer interaction by means of video-based tongue protrusion detection (メディア工学)

概要

論文の詳細を見る
We propose a human-computer communication method based on tongue protrusion. We consider that the normal place of the tongue is inside the oral cavity. Therefore, the appearance of the tongue on the surface of the mouth will communicate the intention of the user to generate an event. This appearance will be detected by our system, by analyzing the changes occurring in the video signal coming from a regular web camera. Three templates are used: a mouth template; and, a left-half mouth sub-template and a right-half mouth sub-template, which were obtained by dividing the mouth template by half. We use the Normalized Correlation Coefficient (NCC) for tracking the mouth region at each frame, and also for comparing the similarity between the new tracked mouth and the original template. When the NCC falls behind a threshold, we assume that the change in the mouth is caused by the tongue protrusion. Then, we analyze both the Left and Right templates NCCs, and we use this similarity measure for detecting whether the tongue was protruded to the left or right side of the mouth. Subsequently, having a positive detection will generate a simple task in the computer. Experiments show that the system is adaptable to different users by being calibrated according to the user's physical characteristics, and that the communication method is easy to learn and perform.
社団法人映像情報メディア学会の論文
2009-10-31