Department of Electrical Engineering Stanford University Yi Liang, Eric Setton and Bernd Girod Channel-Adaptive Video Streaming Using Packet Path Diversity and Rate-Distortion Optimized Reference Picture Selection
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Department of Electrical EngineeringStanford University
Yi Liang, Eric Setton and Bernd Girod
Channel-Adaptive Video Streaming Using Packet Path Diversity andRate-Distortion Optimized Reference Picture Selection
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Motivation
Channel-adaptive packet dependency control: Higher error-resilience; no packet
retransmission Low-latency streaming (<1
second) for interactive applications
[Liang, Flierl, Girod, ICIP’02]
Path diversity Uncorrelated channel statistics Mitigate negative impact on data
This work Low-latency streaming using
multiple paths for further improvement
R-D optimization for multiple streams
1 2
Datatraffic Data
traffic
ContentServer
Client
Relay Server
Relay Server
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Outline
Prior work Channel-adaptive reference picture selection Optimal packet dependency control for multiple
streams Results and video demo
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Prior Schemes
Video redundancy coding (VRC) [H.263++]
Multiple state coding [Apostolopoulos]
RPS-NACK[Lin et. al., ICME’01]
1 3
2 4S S
0 2 8
1 3 5 7 9
4 6 10
NACK (1) ACK (7)
Path 1
Path 2
Path 1
Path 2
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Dependency Control over Multiple Paths
vvopt
vvv
Jv
In
n-vV
vRDJ
min arg
as channel same the
oversent frame
for ,
}{}
|,...3,2{}1{
Path 1
Path 2
Long-Term Memory V=5
n-5 n-4 n-3 n-2 n-1 n
Prediction modes: v=1, 2, … V, I
},5,3,2,1{ I
V=1
,1J
V=2
,2J
V=3
,3J
V=5
,5J IJ
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Path Selection
Path selection: to minimize the distortion of next frame and take advantage of path diversity
Alternate when both channels are good
Send small probe packets over the channel in bad state
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Results (1)
Channel loss_rate_1=loss_rate_2 =15%
LTM len=12Avg burst len=8Feedback delay=6
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Results (2)
Channel loss_rate_1=loss_rate_2 =15%
LTM len=12Avg burst len=8Feedback delay=6
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Video Demo (1)
Proposed scheme RPS-NACK
Mother-Daughter 318 Kbps20% loss
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Video Demo (2)
Proposed scheme RPS-NACK
Foreman 185 Kbps20% loss
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Conclusions
Eliminated the need for retransmission; low-latency (<1 second) streaming over multi-
path
Increased error-resilience by using Packet path diversity Feedback-based path selection Optimal packet dependency control
Improved tradeoff between error-resilience (independent streams) and coding efficiency using R-D optimization
Liang, Setton and Girod: Packet Path Diversity and Rate-Distortion Optimized RPS
Estimate of the Distortion
q
q
pq
p
p
),( 11 DR
),( 22 DR
),( VV DR
…
q
p
D11, p11=q3
D12, p12=q2p
D18, p18=p3
…
…
8
1111
iii pDD
P1
4
1222
iii pDD
D21, p21=q2pq
p D22, p22=up
D23, p23=up
D24, p24=p2
P2n-3 n-2 n-1 n
Related Documents
