A Hardware-Friendly Acceleration of VVC Affine Motion Estimation Using Decision Trees

Digital videos have become ubiquitous across various devices, presenting challenges in transmitting and storing due to their substantial data volume. Video encoders play a vital role in addressing these challenges by improving data compression. The Versatile Video Coding (VVC) encoder, designed to enhance encoding processes compared to its predecessors, introduces a novel tool for Inter-Frame Prediction called Affine Motion Estimation (AME). This paper focuses on the Affine tool inside the Inter-Frame Prediction step, with the aim of developing an optimized, hardware-friendly algorithm. The goal is to reduce computational effort while maintaining the coding efficiency of the VVC encoder. To achieve this, Machine Learning models based on Decision Trees, known for their simplicity in implementation, are employed. The proposed solution successfully achieves an average time reduction of 5.07% in total encoding time, with a minimal coding efficiency loss of 0.23%.