Rate Control with Delay Constraint for H.265/HEVC
Hong Gao, Yuan Zhang · 2021
This paper presents a Reinforcement Learning (RL) based rate control scheme for low latency video communication with High Efficiency Video Coding (HEVC). To avoid buffer overflow and underflow with a small buffer size constraint, we propose a new bit allocation and Quantization Parameter (QP) decision method based on the buffer status to control the buffer occupancy. Different from the heuristics design, the proposed RL-based rate control algorithm uses a neural network to allocate the target bit number and determine the QP value. Experimental results show that the proposed scheme effectively reduces the bit rate fluctuation and can avoid buffer overflow and underflow, which ensures a higher control accuracy and more consistent video quality than other existing methods.