採用以MPEG-4物件形式視訊編碼之視訊會議傳送端之整合

Abstract

在本篇論文中，我們設計並實現一個在個人電腦上的物件形式視訊會議傳送端。並將此系統用來組成一個在個人電腦上的多點視訊會議系統。採用物件形式視訊編碼的最主要理由是為了節省資料量。 此傳送端的架構是由擷取、前置處理、MPEG-4編碼器、即時傳輸規約(RTP)所組成。在影像擷取的控制上，我們引用了VfW的模組來達成；而在聲音擷取的控制上，我們則是引用了MCI模組來達成。 我們在前置處理時則能得到影像切割畫面。這一級的基本概念是將畫面與其相對應的背景相減以得到移動的物件。首先，我們估計攝影機的雜訊，並且把此結果拿來當做往後參數調整的參考。為了消除因為物體內部的平坦區域所造成的錯誤背景，首先，我們先取得一個初步的背景。接著，我們利用影像變化加上填補及收縮的技巧來取得一個粗略的物件輪廓，並利用此資訊來修正初步的背景。 接下來，我們使用一公開的程式Microsoft MPEG-4 Software加以修改以完成編碼與解碼系統。希望能有高速的硬體及有效率的軟體，我們採用了平行處理方式，而用Intel的MMX指令集來實現此方法。 希望我們所採用的網路傳輸協定的主要目的是為了實現多點即時系統，但我們仍能控制所傳送的封包，首先，採用了媒體傳送整合框架，雖然它隸屬於MPEG-4規格的一部份，但並沒有被完善發展。所以我們改採用即時傳輸規約來滿足我們的需求。 最後，在配備Intel Centrino Pentium M 1.5 GHz cpu及512 MB DDR RAM之個人電腦及Microsoft Windows XP Professional作業系統下的測試結果，我們所傳送的影像平均每秒10.7張。

We consider the design and implementation of an object-based videoconference transmitter on personal computer (PC). The purpose is to support multi-point virtual conferencing. The main reason that we introduce object-based video encoder into video conference is saving data amount. The structure of the transmitter system consists of capture, pre-processing, MPEG-4 encoder, RTP (Real-time Transport Protocol). The video capture process is aided by the VfW (Video for Windows package), and the audio capture process is aided by MCI (Media Control Interface) library. We get the segmented image in the pre-processing stage. The basic idea of the stage is a background subtraction technique. First, we estimate the camera noise and the result is used to decide the thresholds. Due to the problem of flat inner regions, we use short-term background to obtain an initial background which usually includes many flat inner regions at first. Second, a temporary foreground mask is obtained to remove the flat inner regions in the short-term background. Next, we use the public-domain software, Microsoft MPEG-4 software, to establish an MPEG-4 coding and decoding system. Hope to use high-processing-speed hardware and effective software to achieve real-time MPEG-4 encoder, we introduce parallel processing which we implied with Intel’s MMX technology into this software. Hope to use network protocol which goal is to realize multi-point real-time system, but we could still control the package we delivery. First, we choose DMIF, Delivery Multi-media Integration Framework, which is belonged to MPEG-4 format, as the protocol. However, the algorithm of DMIF isn’t developed well and is almost given up in MPEG-4 conference. Hence we choose RTP, Real-time Transport Protocol to satisfy the need. Finally, the average frame rate we deliver is 10.7 frames per second on our test system. The test system is based on Intel Centrino Pentium M 1.5GHz, 512 MB DDR RAM and Microsoft Windows XP Professional Version 2002