Exploring the influence of motion estimation algorithm selection and its parameters on the quality of HEVC-encoded 4K drone footage

Real-time video coding on UAV devices presents a challenge where all components of the system must be optimised to get the lowest latency in the video transmission. We focus on the coding and decoding part of the pipeline, where the coding side is located on the UAV and needs to also be energy efficient. Choosing the right number of reference frames which are looked at when searching for a reference block in the motion compensation in High Efficiency Video Coding (HEVC) is crucial for getting the requested video quality and maintaining the latency and power consumption in the requested range. By increasing the number of searched reference frames, the quality increases, but as a drawback, the power consumption and latency also increase. We investigated the impact of the Group of Pictures (GOP) structure and the selection of the Motion Estimation (ME) algorithm on the quality of 4K drone footage encoded using the Kvazaar implementation of the HEVC standard. We vary different encoding parameters such as ME algorithm, ME search step, number of referent frames used in ME, target bitrate, etc. After encoding, three objective quality measures were calculated for all the videos. Since there were over 4000 examined videos, everything was done remotely on a supercomputer BURA.