La lecture en ligne est gratuite
Le téléchargement nécessite un accès à la bibliothèque YouScribe
Tout savoir sur nos offres
Télécharger Lire

Video compression systems for low-latency applications [Elektronische Ressource] / von Ralf Michael Beuschel, geb. Schreier

195 pages
Video compression systems for low-latency applications DISSERTATION zur Erlangung des akademischen Grades eines DOKTOR-INGENIEURS (DR.-ING.) der Fakultät für Ingenieurwissenschaften und Informatik der Universität Ulm von RALF MICHAEL BEUSCHEL, GEB. SCHREIER AUS FRIEDRICHSHAFEN Gutachter: Prof. Dr.-Ing. Albrecht Rothermel Mag. DI Dr. Margrit Gelautz, ao. Univ. Prof. Amtierender Dekan: Prof. Dr.-Ing. Michael Weber Ulm, 22.01.2010 Video compression systems for low-latency applications Abstract: This dissertation presents an in-depth analysis of latency sources in real-time video compression systems and describes strategies to minimize the system delay. Initially, the latencies of intra-frame and predictive coding in standard-mode video compression systems are examined. A theoretical analysis identifies the bitsream buffers – which are a pre-requisite of constant transmission bit-rate – as major sources of latency. In a second step, the results of experiments conducted using the “leaky-bucket” model on ten video sequences with sub-frame-level resolution are discussed. Based on these observations, the intra refresh coding method is identified as a key technique for low latency in predictive video compression systems.
Voir plus Voir moins




Video compression systems
for low-latency applications


DISSERTATION


zur Erlangung des akademischen Grades eines



DOKTOR-INGENIEURS

(DR.-ING.)


der Fakultät für Ingenieurwissenschaften
und Informatik der Universität Ulm


von


RALF MICHAEL BEUSCHEL, GEB. SCHREIER

AUS FRIEDRICHSHAFEN



Gutachter: Prof. Dr.-Ing. Albrecht Rothermel

Mag. DI Dr. Margrit Gelautz, ao. Univ. Prof.

Amtierender Dekan: Prof. Dr.-Ing. Michael Weber



Ulm, 22.01.2010


Video compression systems for
low-latency applications




Abstract:
This dissertation presents an in-depth analysis of latency sources in real-time
video compression systems and describes strategies to minimize the system
delay.
Initially, the latencies of intra-frame and predictive coding in standard-mode video
compression systems are examined. A theoretical analysis identifies the bitsream
buffers – which are a pre-requisite of constant transmission bit-rate – as major
sources of latency. In a second step, the results of experiments conducted using
the “leaky-bucket” model on ten video sequences with sub-frame-level resolution
are discussed. Based on these observations, the intra refresh coding method is
identified as a key technique for low latency in predictive video compression
systems. Finally, a novel approach for intra refresh coding is presented, which
simultaneously satisfies the traditionally contradictory goals of high coding
efficiency and low system latency. Table of Contents

1 INTRODUCTION 1
1.1 Motivation
1.2 Background 1
1.3 Goals and methodology 3
1.4 Structure of the document 4
2 VIDEO COMPRESSION SYSTEMS 6
2.1 Physiological motivation of video compression techniques 6
2.2 Historical development of image and video compression standards 11
2.3 Technical aspects of video compression systems 13
2.3.1 Lossy image compression techniques 13
2.3.2 Principles of hybrid video codecs 15
2.3.3 Basic video coding tools 20
2.3.4 Temporal prediction tools 22
2.3.5 Compressed video stream hierarchy 25
2.3.6 Digital video representation 28
3 VIDEO CODING MODE ANALYSIS 32
3.1 Test Sequences 32
3.2 Definitions and quality metrics 34
3.3 Comparative coding mode analysis 35
3.3.1 Test set-up and quality adjustment 36
3.3.2 Intra coding mode analysis 38
3.3.3 IP-coding mode analysis 42
3.3.4 IPB coding mode analysis 47
3.3.5 Coding mode analysis summary 49
3.4 Localized data rate analysis 51
3.4.1 Macroblock data rate variations 51
3.4.2 Slice resolution and randomized data rate analysis 54


- I - Table of Contents
4 LATENCY ANALYSIS FOR STANDARD-MODE
VIDEO COMPRESSION SYSTEMS 57
4.1 Generalized video transmission model 57
4.2 Algorithmic encoder and decoder latency 60
4.3 Transmission latency 63
4.4 System latency analysis for standard coding modes 67
4.4.1 Intra coding mode 67
4.4.2 Randomized intra coding mode 68
4.4.3 Predictive coding mode 69
4.4.4 Bi-directional prediction modes 71
4.4.5 Interlaced coding modes 73
4.4.6 System latency analysis summary 74
4.5 Experimental buffer latency analysis 76
4.5.1 Leaky bucket model 76
4.5.2 Intra coding mode 81
4.5.3 Predictive coding mode 84
4.5.4 Bi-directional prediction modes 87
4.6 System latency comparison for standard coding modes 88
5 OPTIMIZED INTRA REFRESH ALGORITHM 91
5.1 Introduction to the intra refresh principle 91
5.2 An adaptive region-based intra refresh scheme 95
5.2.1 Region-based forced intra refresh methods 95
5.2.2 Motion-adaptive intra refresh method 97
5.2.3 Selection algorithm for optimal refresh 101
5.3 Real-time implementation of the adaptive intra refresh 108
5.3.1 Frame-level flow chart 108
5.3.2 Structural overview 110
5.3.3 Estimation algorithm program flow 112
5.4 Intra refresh error recovery and scene changes 115
5.5 Low-latency rate control issues 117
5.6 H.264 Implementation aspects 121

- II - Table of Contents
5.7 Intra refresh coding mode latency analysis 124
5.7.1 Frame-CBR intra refresh 124
5.7.2 GOP-CBR intra refresh 128
5.8 Joint data rate and latency optimization 133
5.9 Coding mode performance summary 137
5.9.1 Buffer latency comparison 137
5.9.2 System latency summary 138
5.9.3 Latency, data rate and error recovery comparison chart 139
6 IMPLEMENTATION CONSIDERATIONS FOR LOW-LATENCY VIDEO
COMPRESSION SYSTEMS 143
6.1 Introduction 143
6.2 DSP implementation of an MPEG-2 encoder 144
6.2.1 Detailed functional overview 144
6.2.2 Program-flow optimizations for low-latency encoders 145
6.2.3 Memory usage and access optimizations 147
7 SUMMARY & CONCLUSIONS 152
8 BIBLIOGRAPHY 154
9 LIST OF ABBREVIATIONS 162
10 LIST OF SYMBOLS 165
11 LIST OF FIGURES 166
12 LIST OF TABLES 169
APPENDIX A – H.264 AND MPEG-2 QUALITY ADJUSTMENT 171
APPENDIX B – SIMULATION RESULTS 173
APPENDIX C – INTRA REFRESHALGORITHM FLOWCHARTS 176
APPENDIX D – INTRA REFRESHEXCESS DATA RATE SIMULATIONS 178
APPENDIX E – INTRA REFRESH TABLE INITIALIZATIONS 180
APPENDIX F – ENCODER CONFIGURATION 186
- III -


- IV -
1 INTRODUCTION
1.1 Motivation
Controlling and minimizing the latency of real-time video compression systems is
becoming more and more critical as this technology is employed in applications
which traditionally use uncompressed, latency-free video transmission. With a
focus on video broadcast and video conferencing, contemporary video
compression standards were not designed to meet the latency requirements of
emerging applications such as automotive driver assistance cameras, remote
controls or RF studio cameras. Especially if high compression ratios must be
achieved, the common techniques introduce unacceptable buffer delays to achieve
a constant transmission bit rate. Therefore these new applications can only be
realized if the compression tools and algorithms as well as the target hardware are
optimized for minimum latency.
Designing systems for low-latency video transmission requires basic
implementation knowledge as well as profound knowledge of the video coding
algorithms and buffer management. Video compression standards only describe
the algorithmic decoding procedure and profile features which ensure the
interoperability of different encoders and decoders. Nevertheless, the system
designer has many options in selecting coding modes, coding parameters and
buffer sizes to meet the requirements of a specific target application.
1.2 Background
Modern video coding applications are based on the concept of hybrid video
coding, which is a combination of lossy intra picture coding and (principally
lossless) inter picture prediction. As bandwidth was not a major concern in
classical video broadcast applications, the video compression technology was
initially driven by video conferencing applications. The enabling technology for
efficient video compression is high-performance digital signal processing, because
improvements in the coding efficiency are mostly achieved at the cost of increased
computational complexity.
Since the first digital video coding standard released in 1984, it has been possible
to improve the efficiency continuously because of the decreasing cost of
computing performance. Furthermore, the progress in storage and communication
technology has enabled higher data rates, allowing to continuously increase the
image resolution and visual quality over the last 20 years.
In the course of the last two decades the video coding technology was mainly
driven by four applications, namely:
- 1 - Chapter 1 Introduction

video conferencing and video telephony
video storage and retrieval, e.g. video archives, video-CD, DVD (digital
versatile disc)
video broadcast, e.g. DVB (digital video broadcast), DMB (digital media
broadcast)
(Internet) streaming

Due to the different requirements of these applications, a significant number of
standardized and proprietary video coding techniques were established.
Furthermore, within all relevant standards, the coding parameters can be adjusted
to match the characteristics of different applications.
For the purpose of storage and retrieval, the compression efficiency is usually the
major concern to achieve the highest possible quality with a minimum storage
cost. Hence, these systems use the most advanced algorithms which require the
highest computational performance, backward temporal prediction and possibly a
pre-analysis of the complete video clip. This is, however, not a critical issue if the
encoding can be done offline without real-time constraints of resources and
algorithms. The characteristics of archived video streams are generally
contradictory to low-latency transmission and processing.
Live broadcasting and streaming applications impose certain real-time constraints
on the system which limit the choice of algorithms as it is not possible to perform a
pre-analysis of the complete video. In the case of video broadcast servers, the
computational complexity is a minor issue since initial hardware costs pay off
quickly against broadcast transmission bandwidth. The high coding efficiency of
broadcast applications is achieved with the same coding tools as used for storage
applications but with a short pre-analysis interval of a few frames. This results in
processing, buffer and transmission delays in the range of multiple seconds.
In the case of video conferencing and video telephony applications, the
requirements differ significantly. The major issue here is the reduction of latency
to an end-to-end delay of 100 ms which is required for unimpaired conversation.
Therefore, the video transmission must achieve the same delay to guarantee the
lip-sync of audio and video. Limitations on computational resources in mobile and
relatively cheap consumer devices also restrict the choice of algorithmic tools
resulting in reduced coding efficiency. Furthermore, the point-to-point
characteristics of video conferencing impose restrictions on the bandwidth
because the transmission capacity is paid by only two users rather than a large
broadcasting audience.
- 2 -

Un pour Un
Permettre à tous d'accéder à la lecture
Pour chaque accès à la bibliothèque, YouScribe donne un accès à une personne dans le besoin