Current Implementation
A screenshot of running GColl videoconference
A multi-point conference must be established to set up the GColl environment for multiple remote participants. As multicast is not widely available yet, we have decided to use the RUM2 packet reflector (developed in our lab) instead, which implements multicast functionality at the application level. From the network point of view, scalability of this solution is lower than it is on multicast; on the other hand, packet reflector provides us with various data processing capabilities. RUM2 capabilities were exploited to implement the gaze awareness in GColl. A single stream of video is chosen for each participant to be sent to other users of the environment: focus camera stream is sent only to the user in the sender’s focus (if any); camera stream is sent to all others. This approach also saves a lot of bandwidth at participants’ workstations — approx. 35% for 6 users when compared to multicast. Since reflector sends only a single stream per participant to each user, no further logic is needed to process the video received on the client side. We have used Vic (a part of UCL Media Tools) as a base for the implementation of the design on the client side. A number of new functions was implemented so that Vic could be used as a videoconferencing tool in the GColl environment. These are especially automated visualisation layout of the videostreams, switching streams in the emph{focus} window, and client-client and client-reflector communication. Although the GColl design has no inherent limitation on number of sites participating in the conference, the current implementation of automated videostreams layout supports two sites only — the work on the support of multiple sites is currently under way.
Currently we are focusing on turning the prototype implementation into more stable version. As the underlying videoconferencing client is not actively developed at the moment, we are reworking some parts of its code to better comply with currently available technologies. Microsoft Windows version of the client is under the development, too. Functionality of the client is being further improved in a these ways:
- Head position detection. The camera stream will be acquired in higher resolution and smaller frames will be cropped from it before sending to the network. These resulting frames should have the user’s head positioned in the center of the cropped image. This will allow a GColl user to move in front the camera more freely.
- Arbirtrary image scaling. Original client supports rescaling of received videostreams in multiples of two only. We are reworking the video rendering in the GColl client to allow arbitrary scaling ratio, which will allow better fitting of videostreams on the GColl screen.
