4K-UHD real-time HEVC encoder with GPU accelerated motion estimation

摘要

This paper proposes an efficient implementation of an HEVC encoder with highly parallelized motion estimation on GPUs. It is an obstacle to parallelization of conventional motion estimation that the processing results of the neighboring blocks are used for efficient data compression. This paper especially proposes two new methods on the encoder for relaxing the data dependencies by dividing the motion estimation process into multiple steps and using the results of the previous steps, instead of the results of neighboring blocks in the same step. This allows the full use of many processor cores on GPUs while maintaining compression efficiency. The experimental results show that the proposed encoder achieves 60 fps real-time encoding for 4K-UHD sequences and 10x speed-up with an acceptable small bit-rate increase compared to the x265 encoder.