GPU accelerated foreground segmentation using CodeBook model and shadow removal using CUDA

Improving the Performance of Video Processing on Hadoop Clusters

“…11 shows that SVSwFW outperforms both NVSwO and SVSwS. This is because if the Splitting occurs at a frame crowded with slow-moving foreground objects, the recent history is more likely crowded with the same foreground objects, especially when a recording with a high frame rate is considered.…”

“…On GPU architecture, there are many implementations to parallelize BS using di↵erent optimization techniques (such as memory coalescing, data transfer, kernel overlapping, divergent branch elimination, and e cient register usage) [8][9][10]. The research described in [11] proposed a parallel implementation of the CodeBook model on GPU to achieve BS. On distributed memory systems, the authors of [12] proposed a parallel algorithm of the classical Gaussian model to support real-time video applications depending on CD.…”

“…11 shows the e↵ect of two-node failure on the performance of Default-DPP and ECW-DPP, where three replicas are used to support the recovery. As demonstrated in the figure, there is no significant di↵erence between the two DPP.…”