Information Technology Reference
In-Depth Information
In their work on interactive streaming of light fields, the authors predict the ( x , y )
mouse co-ordinates based on dead reckoning and translate these into the view-
point [41]. The use of a Kalman filter for head movement prediction in scenarios
where head movements can control the application have been proposed in [42]. In
prior work on dynamic light fields, six Kalman filters have been used for predict-
ing the 3-D co-ordinates and the 3 Euler angles that define the viewpoint [43, 44].
The viewpoint and the rendering algorithm together determine the number of views
that need to be streamed to the client. The authors mention two possible system
design choices. Viewpoints exceeding the bit-rate threshold can be disallowed or
those viewpoints can be rendered with lower quality by not streaming all the views
demanded by that viewpoint. The authors also note that if the streaming system al-
lows tuning into a view-stream only during certain frame-intervals, one can choose
an appropriately long prediction lookahead and tune into new view-streams before-
hand to avoid missing the join opportunities.
2.3
Multicasting
Multicasting can drastically reduce the bandwidth required from dedicated media
servers. IP multicast, specified decades ago [45], allows sending an IP datagram to
a group of hosts identified by a single IP destination address [46]. Hosts may join
and leave a multicast group at any time. This requires multicast-capable routers that
replicate packets as required. Even though IP multicast is extremely efficient at dis-
tributing data to multiple interested receivers, most routers on the Internet keep this
functionality turned off due to reasons related to security, billing and the size of
the data-structures to be maintained by the router. Nevertheless, the bandwidth con-
servation benefits of IP multicast have resulted in rising deployment for corporate
communications and, more recently, IPTV service.
The seminal work on receiver-driven layered multicast (RLM) [47] focuses on
video streaming without interactive pan/tilt/zoom. The authors propose compressing
the multimedia signal in hierarchical layers and letting individual receivers choose
the layers to join. Receiving more layers leads to better quality. Each layer is deliv-
ered using a different multicast group. Note that if a receiver joins too many layers
and creates congestion on a link then packets can be dropped indiscriminately from
all layers affecting received quality, possibly for multiple receivers that share the
congested link. A receiver performs regular tests to decide if it should unsubscribe
already joined layers or subscribe new layers. “Shared learning” among receivers
can reduce the number of tests and hence the convergence time.
Recently, the RLM framework was adapted for interactive dynamic light field
streaming [43]. Depending on the chosen viewpoint, the client decides which views
and consequently which multicast groups to subscribe. The latency for joining a
new multicast group is generally low with IP multicast [48]. As in the case of RLM,
it is the client's responsibility to avoid congestion on intermediate links. The source
does not adapt transmission to curtail congestion; it keeps transmitting IP datagrams
to the multicast groups' addresses.
 
Search WWH ::




Custom Search