3D Scene Streaming over P2P Network
Date Issued
2007
Date
2007
Author(s)
Huang, Ting-Hao
DOI
en-US
Abstract
三維模型與場景的資料量日益龐大，為了將之傳送於網際網路上，壓
縮與串流兩種方法是勢在必行的。因此，我們將三維模型的各項資
訊，以參數化的方式轉換成二維影像，並使用JPEG 2000 影像壓縮標
準來達到無失真的壓縮，可以達到約33%的壓縮率。所以三維在傳輸
三維模型時，便可利用JPEG 2000 的種種特性，包括：Progressive
Transmission、Region of Interest 等，來達到三維模型串流的效果。此
外，為了在點對點的網路架構中傳輸三維場景資料，我們將JPEG 2000
的資料依解析度以及區域先作切割，並計算每個物件的視覺重要性，
以決定每一段資料的傳輸順序。實驗結果說明本系統可以在使用較少
的頻寬下，達到三維場景傳輸的目的，並將視覺上的缺陷減到最少。
As the data size of 3D models increased, the data size of 3D scene also increased. The most relative example is 3D online games. The data size is so large that most of the game data
are shipped using high capacity DVDs. However, to browse a large virtual world online, downloading the entire data is no option. In fact, only a small portion of the world is needed
to render an acceptable result since most objects in a scene are to far to be seen, or hide behind opaque objects. To avoid the long waiting time of downloading the entire scene before browsing the virtual world, the concept of ”3D Scene Streaming” is proposed. Its general idea is: given the information of the scene and the position and viewing direction of the user, only objects which are seen, and will be seen in the future, are needed to be downloaded.
The major question is to decide the priorities of the objects to be downloaded.
To the best of our knowledge, there're two approaches to achieve 3D scene streaming. The first one is view-dependent based approach, which considers the position of user, his/her
viewing direction, and the distance between the user and object to decide the precedence of downloading 3D objects. The other approach is visibility preprocessing, which divide the entire scene into multiple cells and find the “potential viewable space” of any pair of cells. The former is quite suitable for outdoor scenes, but its efficiency may drop in indoor scenes because many objects near to user but occluded by opaque object are possible to be transferred earlier. On the other hand, the latter is very suitable for indoor scenes since architects usually designed following some rules such as symmetric.
In this thesis, we adapt the view-dependent approach to achieve 3D streaming. In other words, we find the visual importance of each 3D object to decide their order of transmission. To address the scalability problem of traditional client-server network architecture, we
separate 3D object into data pieces and apply FLoD for peer-to-peer network architecture. Experiment result shows that our system can display 3D scene with lowest quality loss under limited bandwidth, and the work load of server is reduced at the same time.
Subjects
三維場景串流
同儕網路
3D Scene Streaming
Peer-to-peer
JPEG-2000
Type
thesis