Speed-up optical flow calculation using graphical processing units

Nowadays, optical flow motion vectors are using some areas like moving object detection and tracking with successfully. But computation of these motion vectors on central processing units (CPUs) is quite slow, because each point, which is taken from image, has to calculate separately. In this paper, optical flow motion vector computation became parallel using graphical processing units (GPUs) and observed improvements on computation time are mentioned. Also these explanations are supported with software and hardware perspectives. In experimental results, described optical flow models are visualized to check generated motion vector results accuracy and these models computation time on GPU and CPU units are investigated comparatively.