A Study of Fast Software-Based Video Computing for Real-Time Multimedia Applications

瀏覽統計 Email 通知 RSS Feed

簡歷

基本資料

Project title

Code/計畫編號

NSC98-2221-E019-036-MY2

Translated Name/計畫中文名

應用於即時多媒體應用之快速軟體為主的視訊計算方法研究

Project Coordinator/計畫主持人

Shyi-Chyi Cheng

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=2010401

Year

2010

Start date/計畫起

01-08-2010

Expected Completion/計畫迄

01-07-2011

Bugetid/研究經費

567千元

ResearchField/研究領域

資訊科學--軟體

"快速發展的網際網路引進包括視訊資料在內之大量多媒體資訊在網路上傳送，視訊資料處理的高計算複雜度特性限制了在網際網路建構即時視訊應用系統的可能性，常見的效能改善方式是運用超大型積體電路加速高計算複雜度的即時視訊資料處理模組，然而一些重要的視訊應用系統如H.264/AVC編碼系統及視訊檢索系統往往不具備細微平行化因子，因此純硬體化的加速方法的效果往往受到極大的限制。本計畫捨棄硬體化的加速方法轉而設計軟體為主的視訊系統加速方法，其原因在於軟體為主的視訊處理架構更適合用於設計快速之可調適視訊應用。快速軟體為主的視訊處理架構的核心為計算察覺技術，計算察覺技術基於一計算-失真最佳化方法尋找影響視訊處理架構效能的關鍵工作，並優先執行具有最大效能改善效果的步驟。具有即時限制特性的視訊應用往往無法保證當程式執行時間用盡時，系統已達最佳狀態，計算察覺技術優先執行具有最大效能改善效果的步驟的特性可有效解決這個問題，因此具計算察覺特性的視訊應用系統往往可以在系統尚未檢驗所有可能解之前的情況下就已達一定程度的系統效能。本計畫為一三年期計畫，第一年，我們將提出一具計算察覺特性的即時軟體式的H.264/AVC編碼架構，H.264/AVC需要大量記憶體存取頻寬的問題也因引進區塊索引結構而獲得解決，區塊索引技術提供本視訊編碼系統內容導向記憶體隨機存取能力，具計算察覺特性的即時軟體式視訊處理架構開創了設計即時H.264/AVC編碼系統的新方法。本計畫的第二年，我們投注精神在基於3維視訊資料向量量化技術的視訊立方格為主的檢索系統設計，一個視訊片段首先被切割成多個畫面群，每個畫面群更進一步被切割成多個視訊立方格，然後我們使用3維矩量保持技術抽取每個視訊立方格內包含時間-空間的重要特徵，進而我們設計了一個視訊片語意標示系統。最後，本計畫的第三年，我們擴展第一年所使用的區塊索引技術為視訊立方格索引技術，應用第一年所發展的基於索引架構的計算察覺技術，我們設計了一個具計算察覺特性的即時軟體式視訊檢索系統，同時我們也提出使用PS3細胞式多重處理器平行化處理本計畫所提出之視訊檢索系統，初步實驗結果驗證本計畫提出的方法的可行性。" "The rapid advance of the Internet leads to the mass multimedia information including video data transmitted through the network. The high computational complexity of video processing limits the real-time video-based applications over the Internet. The general technique to facilitate real-time video processing is to parallelize the tasks of video computing and implement those tasks using VLSI hardware circuits. However, some important video-based applications such as H.264/AVC video coding and video retrieval do not inherit fine-grain parallelism in their architectures. This limits the efficiency of video processing using pure hardware solutions. Instead, in this project, we pay attention to speeding up the problems of video computing using a software-based approach which is more suitable to design fast adaptive algorithms for video-based applications. The core of the fast software-based video computing is the computation awareness, which looks for critical tasks based on a computation-distortion optimization scheme. The deadline of a real-time video-based application often limits the efficiency of the system since the obtained solution is not guaranteed to be optimal when the system stops. To solve the problem, the tasks that would bring significant contributions to the efficiency and effectiveness of the system should be performed first. This guarantees the system performance to a certain degree even when not exploiting all feasible solutions. In this project, three challenged problems of video computing are studied using the proposed fast software-based approaches. The schedule of the project is separated into three years. In the first year, we focus on the design of a computation-aware H.264/AVC using content-based image indexing techniques. The problem of large amount of memory accesses in H.264/AVC is properly solved by the indexing structure which offers effective random access capability to the video coding system. As compared with the hardware speedup methods, the proposed method offers much fine-grained parallelism which is more suitable to implement in parallel. The computation-aware characteristics of the method span a new way to design a real-time H.264/AVC system. In the second year, we focus on the design of a cube-based video abstraction using 3-D vector quantization. An input video shot is first separated into multiple frame groups, where each of them is further divided into multiple non-overlapped video cubes. The spatial-temporal feature of a video cube is extracted analytically using the 3D moment-preserving technique. The use of labeled and unlabeled data to annotate video content is also proposed in order to summarize database videos into a hierarchical representation according to user preferences. Finally, in the last year, we aim at extending the block indexing into video cube indexing in order to design a fast computation aware video retrieval system. Furthermore, a parallel implementation for the proposed video retrieval scheme based on PS3 cell processors architecture is described. Preliminary experimental results show the feasibility of the proposed methods."

Keyword(s)

視訊編碼
H
264/AVC
計算察覺運動偵測
平行處理
內嵌式系統
視訊檢索
3維矩量保持技術
視訊立方格分析
速率-失真度-計算最佳化
視訊資料索引
Video coding
H
264/AVC
computation aware motion estimation
parallel processing
embedded system
video retrieval
3D moment-preserving technique
video cube analysis
rate-distortion-computation optimization
video indexing

DSpace CRIS

A Study of Fast Software-Based Video Computing for Real-Time Multimedia Applications

基本資料

Description