Contents of this page are obsolete. This page is preserved and stored at this URL just from historical viewpoint. Original URL was http://www.mm.media.kyoto-u.ac.jp/members/kameda/...
Please visit www.kameda-lab.org for recent information. (2002/12/06, kameda@ieee.org)

Publication, Research, Docs and Info, KAMEDA


next up previous
Next: Experiment Up: 3D Reconstruction Method Previous: 3D Reconstruction

Pipeline Processing

Our 3D reconstruction system named SCRAPER can be divided into three stages.

  1. image capture
  2. extraction of dynamic region
  3. ESS calculation by VFM method
Let us call this sequence of stages a path. If the 3D reconstruction is done in this order sequentially, some parts of the system always idle. For example, when images are being captured, extraction and ESS calculation cannot be done. As a result, throughput is low and that is not desirable for real-time VR applications. To improve the throughput, we propose to activate several paths simultaneously in the pipeline architecture. We prepare three kinds of processes: image captor, extractor, and 3D composer and increase number of these processes to support multiple paths.

In our prototype system, an image is captured by the video capture card for which CPU power is not necessary whereas an extractor needs CPU power because it extracts by detecting regions where the pixel values differ from its background image taken beforehand, so the two processes need only one CPU to work together. In addition, captured image data which is transferred to the extractor is not small and so it is not desirable to use physical network device to transmit the data between them. Therefore, We assign one video image captor and one extractor on the same workstation. As a result, the number of video image captors and that of the extractors are the same as that of the cameras.

On the contrary, the number of 3D composers can be increased because the calculation on the 3D composer is completely localized. The system can improve the throughput by preparing the 3D composers on different workstations distributed in a LAN.

As a consequence, the throughput is improved by preparing the multiple paths in the pipeline architecture, which means temporal division of 3D reconstruction process. The number of the paths are subjected to the number of the 3D composers the system can offer. Figure 1 shows the process timing chart when the system has four cameras and four 3D composers and assigns two 3D composers at each path.

We introduce a process named scheduler to synchronize the processes in the pipeline architecture.

  
Figure 1: Pipeline Architecture



next up previous
Next: Experiment Up: 3D Reconstruction Method Previous: 3D Reconstruction



Yoshinari Kameda
Mon Sep 21 11:42:41 JST 1998