An optimal many-core model-based supercomputing for accelerating video-equipped fire detection

摘要: Automatic fire detection has become more and appealing because of the increasing use video capabilities in surveillance systems used for early fire. However, its high computational complexities limit real-time applications. To meet processing today's techniques, this study proposes a single instruction, multiple data many-core model. design an efficient model image applications such as detection, key parameter is data-per-processing-element (IDPE) variation system, which amount directly mapped to each element PE. This quantitatively evaluates impact IDPE on system performance energy efficiency multi-stage approach that consists movement-containing region color segmentation, feature extraction fires, decision making if there or non-fire frame. In study, we six ratios determine optimal provides most operation using architectural workload simulation. Experimental results indicate achieved at 64 value terms worst-case execution time efficiency. addition, compares configuration with commercial graphics unit (Nvidia GeForce GTX 480) show improved proposed algorithm. outperforms graphic