Abstract:
With the growing trend of technology in network and computers science arena, many of the high end applications like multimedia related applications are becoming algorithmically complex and data intensive. Video Surveillance is one such growing business application due to its concern in various areas like commercial security, and military applications. With increasing deployment in large scale distributed environment, the challenge is to process huge multimedia stream data with several computation intensive algorithms. Hardware researchers are trying out to find a cost effective solution for the application. A dedicated hardware solution for the application proves cost expensive and decreases scalability. Alternatively, we have HPC solutions such as cluster, grid and multicore processor for the problem. Research community either had addressed solution for Video Surveillance or for a generalized HPC solution for high end applications but none had addressed solution concerning with video surveillance specific requirements.
In this dissertation work, we present a method to parallelize and implement video surveillance algorithm on the STI Cell Broadband Engine. The methodology provides a conceptual architecture for parallelizing applications with requirements similar to video surveillance. We then briefly present our previous work on computer cluster, discuss various issues and challenges related to porting the code, followed by results demonstrating the speed up and comparison of the results of both approaches.
The implementation of programs is done using C, simulation in Cell SDK 2.0, images used were stored on PC, of Windows bitmap format
