Morphing Parallelization Strategy to Support On-the-Fly Video Analysis

To automate and accelerate Internet video search, we propose a parallelized on-the-fly video analysis that is distinguished from the conventional approaches in the following two features. One is to morph its parallelization by first screening out hundreds of videos at each processor in a low quality of analysis, then repeating this screening process as increasing the analyzing quality, and finally examining each of the final candidates among multiple processors. The other is to download and handle any portion of a video clip independently at a different processor, which is differentiated from the conventional stream-based analysis that scans a video from beginning to end. We have evaluated the promising performance of our proposed system by parallelizing the Auotonoesis colorschema-based video analyzer on top of the AgentTeamwork parallel-computing middleware. This paper presents implementation techniques and preliminary performance to support our parallelization strategy.

• Type: Journal
• Year: 2010
• Where: PDPTA
• Authors: Munehiro Fukuda, Shuichi Kurabayashi, Jeremy K. Hall, Yasushi Kiyoki
• Keywords: Conventional Stream-based Analysis, Distributed And Parallel Computing, On-the-fly Video Analysis, PDPTA 2010, Video