In the usual systems, where each co-located subgroup shares a videoscreen/display (such as most of the commercial products — LifeSize, Tanberg etc.), a great deal of non-verbal signals is lost. One of the most important missing signals is gaze awareness, that is the ability to perceive who is looking at whom. In face-to-face communication we use this information to structure our speech — we do not need to use names to address someone (we just look at him/her), it can be used to signal that someone wants to start speaking etc. Implementation of gaze awareness into a videoconferencing environment is quite tricky, especially if one wants to create a flexible and low-cost system. Among other things, the videoconferencing system has to supply each user with her own visualization device as otherwise those users, who are sharing a device together, are in a sense sharing the “same eyes”– either all of them experience a feeling of eye-contact with a remote participant, or none does. In our system, we have decided to assume, that each user has his/her own notebook, which will act as the individualized visualization screen.

Basic Structure:

A basic structure diagram of a GColl installment on a site with three users is depicted at Figure 1. An image of the whole group is recorded by a single camera G. An echo-canceling microphone is used to capture the audio on this site. Additionally, each user uses her notebook and two common web cameras: the focus camera is attached to the top of the notebook screen itself while the side camera is placed on the desk next to the notebook. If available, a shared screen can be used for slides or any other material the group needs. A scheme of GColl visualization screen and an photo of from our prototype implementation is shown in Figure~\ref{fig:scheme-desktop}. The screen is divided into three parts: the top left image provides a “whole room” view of the remote group (or split into subframes if more than two groups are communicating); images of other participants (remote as well as co-located) are shown on the bottom of the screen and after clicking on any of these small images, the chosen stream is then shown in higher quality in the top right (focus) frame. A click on the whole group frame itself unselects the chosen stream once it is not needed.

Functionality:

The focus frame allows each participant to easily select some other user, on whom she wants to currently focus, and see him in more detail. The system informs other users whether or not they are “in focus” of each of the other participants by selecting the appropriate videostreams to be shown on the bottom of their visualization screen: those users, who are not selected, receive the side camera image; the selected user receives the stream from the focus camera. For example, the user, whose screen is depicted on Figure 2, knows that she has been selected by the participants A, C and F, and that she herself focuses on the participant A. This setup allows us to mediate a sensation of mutual gaze. It is based on findings from cognitive psychology which show that people would still perceive eye-contact if the other person’s gaze is directed less than 5° below the camera. By having a fixed place for the focus window in GColl, it is easy to attach the focus camera appropriately close to the visualized eyes. Thus, if you are selected by another participant and that participant looks at you in his focus frame, you will perceive a feeling of eye-contact. See an illustrative screenshot at current GColl implementation page.