[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
This report describes a method of eye-contact in profile that realizes a face-to-face dialogue on a tele-conference.
In comparison with a real world meeting, there is a limitation of the number of available media to communicate on the tele-conference. Because of this limitation, two factors are important on the tele-conference.
The first factor is eye-contact. If the dialogists(who are the speaker and the listener) make eye-contact, the participants can easily identify the person to whom the speaker talks. Therefore, eye-contact is important for the participants. On the other hand, the dialogists should make eye-contact in order that the speaker knows whether or not the listener shows interest in the dialogue.
The second factor is confirming the self image. The participants should make sure whether or not they are in the view field of the cameras. Therefore, it is important for the participant to confirm his own image that are sent to the other participants. When the speaker talks to the listener, it is much important for them to confirm their own images.
In the previous tele-conference system, a participant cannot make eye-contact with another participant because the position of the camera that takes his image is not in front of his face.
Our method realizes the two factors on the low-cost hardware as follows. The profiles of the dialogists are taken by the cameras at their sides, and the left profile of a dialogist and the right profile of the other dialogist are horizontally aligned on each display.
The participants identify the listener, according to the human behavior. Usually, they look the eye direction of the speaker in order that the participants identify the listener to whom the speaker talks. But if they do not notice the eye direction of the speaker, they estimate it from the direction of his face. When the profiles of the dialogists are horizontally aligned on the display, the direction of each profile shows the eye direction and the dialogists look as if they make eye-contact on the display. Moreover, the dialogists can confirm their own images because their profiles are always displayed.
Because this method applies not to all participants but to the dialogists alone, it realizes a face-to-face dialogue regardless of the number of the participants.
We designed an experimental tele-conference system, and conducted several experiments. In the first experiment, we hold a conference on this system and asked the participants about eye-contact between the dialogists. All participants answered that the dialogists looked as if they make eye-contact and they can easily identify the person to whom the speaker talks.
In the second experiment, we asked a speaker about eye-contact between the listener and him in the dialogue on this system, in order to evaluate that he knows whether or not the listener shows interest in the dialogue. As a result, the speaker knows a degree of interest of the listener.
Finally, we asked the participants about the effect of displaying their own images. They showed in favor of our methods.