https://hgpu.org/?p=5070
A parallel mapping of optical flow to Compute Unified Device Architecture for motion-based image segmentation