Extracting Dynamic Information from Video in Real Time

Abstract

Graphic processing units (GPU) play a major role in providing faster runtime with its parallel architecture. With hundreds of cores, GPUs help in accelerating the runtime of algorithms in various applications such as machine learning and image processing. Also CUDA parallel platform lets us to use CUDA enabled GPU for accelerating the runtime of this algorithm. In this work, we exploit the parallel computing capacity of GPU which will be programming in CUDA in video surveillance cameras. A fundamental task in video surveillance cameras is to capture sparse moving objects (foreground) in slowly changing background. A recent algorithm known as the Prac-Re-Pro-CS appears to be especially appealing from the performance viewpoint. However, the algorithm involves manipulations of large matrices and hence is very computationally intensive. The goal of this work is to achieve a real-time implementation of the PracReProCS algorithm by exploiting the parallel architecture of GPU.