分散並列計算機による実時間ビジョン  [in Japanese] Real - time Vision on a Distributed Parallel System  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

分散並列計算機の1 種であるPC クラスタを利用し,実時間並列画像処理アプリケーションを構築するためのプログラミング環境としてRPV が提案されている.これは,実時間並列画像処理に必要な実時間データ転送機構,同期機構,例外処理機構を提供することにより,アプリケーションを容易に構築することを目指したものである.しかし,RPV の問題点としてレイテンシが大きくなってしまうことがあげられた.そこで我々は,この問題点を解決するために,ストリームデータ転送方式をRPV に導入することを提案する.これは,1 フレーム分よりも小さくデータを分割し,それを単位にデータ転送やデータ処理を行う方式である.本論文では,まずRPV の概要を述べ,その問題点をあげる.次にストリームデータ転送方式について述べ,さらにこれを利用したアプリケーションによる性能評価を行い,有効性を示す.RPV, which is a programming environment for real-time parallel computer vision on a PC cluster, has been proposed. It provides functions of real-time data transfer, synchronization mechanism and error handling to make it easy to construct real-time parallel computer vision applications. The problem of RPV is that the latency becomes long because of frame-based pipeline structure of the system. To make the latency shorter keeping high throughput, we have introduced stream data transfer mechanism, in which one frame data is divided into small packets and data transfer and data processing are applied to each packet. First in this paper, we discuss RPV and its problem. Secondly, stream data transfer mechanism is de-scribed and experimental results using a large-scaled vision application are shown to evaluate the stream data transfer.

RPV, which is a programming environment for real-time parallel computer vision on a PC cluster, has been proposed. It provides functions of real-time data transfer, synchronization mechanism and error handling to make it easy to construct real-time parallel computer vision applications. The problem of RPV is that the latency becomes long because of frame-based pipeline structure of the system. To make the latency shorter keeping high throughput, we have introduced stream data transfer mechanism, in which one frame data is divided into small packets and data transfer and data processing are applied to each packet. First in this paper, we discuss RPV and its problem. Secondly, stream data transfer mechanism is described and experimental results using a large-scaled vision application are shown to evaluate the stream data transfer.

Journal

  • 情報処理学会論文誌コンピュータビジョンとイメージメディア(CVIM)

    情報処理学会論文誌コンピュータビジョンとイメージメディア(CVIM) 43(SIG11(CVIM5)), 1-10, 2002-12-15

    Information Processing Society of Japan (IPSJ)

References:  9

Cited by:  8

Keywords

Codes

  • NII Article ID (NAID)
    110002711533
  • NII NACSIS-CAT ID (NCID)
    AA11560603
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    1882-7810
  • NDL Article ID
    6409564
  • NDL Call No.
    Z74-C192
  • Data Source
    CJP  CJPref  NDL  NII-ELS  IR  IPSJ 
Page Top