Highly-parallel HVEC motion estimation with CUDA [title missing from article PDF]

This article´s PDF file is not fully compatible with IEEE Xplore®. The current document is the best copy available and represents the article as originally published even though it may display with missing fonts, text, or figure elements. No better copy is available from IEEE. IEEE Xplore® will continue working to obtain a better replacement document but we can not guarantee a better copy can be provided. We regret any inconvenience.The High Efficiency Video Coding (HEVC) standard is the most recent standardization work of the Joint Collaborative Team on Video Coding (JCT-VC). It offers a notably better compression performance compared to preceding standards. However, the speed of the encoder software is far too slow to meet the demands of real-time applications. At the same time, the computation capabilities of modern graphics processing units (GPUs) are increasing constantly. In this paper, we present a highly parallel implementation of the HEVC motion estimation (ME) algorithm on the Compute Unified Device Architecture (CUDA). Our approach utilizes one or two GPUs to perform ME, while all other parts of the software stay unchanged. Experimental results show that this hybrid-encoder, manages to compress high-definition video sequences up to 42% faster than the unmodified encoder software, with negligible impact on bit-rate and statistical video quality.