Accelerating HEVC Motion Estimation Using GPU

The demands of faster video streaming and higher resolutions induce the next generation video coding standard, High Efficiency Video Coding (HEVC), which utilizes more complicated codec structures to obtain better video compression ratio than H.264/MPEG4. However, its computational complexity also grows significantly. The paper investigates the acceleration methods of HEVC on Graphics Processing Unit (GPU). The focused kernel is motion estimation since it is the performance bottleneck. We developed a new search pattern which incorporates GPU and CPU, and applied several performance optimization techniques on GPU. Experimental results show that our CPU/GPU implementation can achieve up-to over 17 times speedups, comparing to the state-of-art CPU implementation. For the motion estimation alone, nearly 200 times speed-up can be obtained, with slightly compression bit-rate loss.