Abstract:
Video surveillance systems have found widespread use in applications ranging from homeland security for human lives and property, traffic monitoring, law enforcement practices and military defense. The scientific challenge is to devise and implement automatic systems able to detect, and track moving objects, and interpret their activities and behaviors. The other issues that have come up are in terms of efficiency of the algorithms involved in video surveillance. Issues like robustness, scalability and real time processing are hindering the progress of video surveillance.
The onset of affordable multiprocessors has triggered a shift from clusters to programming using multiprocessors. There is a great enthusiasm in the industry as well as the academic community about the change parallel programming is bringing in. The idea with parallel programming is to use the massive performance given by these multiprocessors to solve the algorithms of video surveillance in real time. Alternatively they can be used to work on video streams of better resolution as well.
We have implemented video surveillance algorithms in a way to reduce the amount of time that is taken to process one frame. The implementations include a intuitive fusion algorithm. We have shown the applications of support vector machines to solve background modeling. We have also implemented background modeling, Embedded Zero Tree Wavelet algorithm, morphological operations and connected components labeling on the GPU and achieved considerable speed up. The idea was to find out the applicability of the GPUs in the field of video surveillance.
