With the explosive growth of video data， limited network bandwidth and high computational demands pose significant challenges for video transmission and storage. In this context， the continuous development of efficient video coding methods is of critical theoretical significance and practical value， as it ensures the delivery of high-quality video services under resource-constrained conditions. However， traditional hybrid video coding frameworks have gradually reached performance bottlenecks， making further improvements in coding efficiency increasingly difficult. In recent years， deep learning， with its powerful nonlinear fitting and representation capabilities， has provided new opportunities for optimizing video coding. This paper presents a systematic and detailed analysis of deep learning-driven video coding technologies. First， we briefly introduce video coding techniques under conventional coding frameworks and further explore the optimization of key modules， such as intra- and inter-frame prediction， through deep learning. Then， we focus on the development and key technical routes of end-to-end video coding frameworks based on deep learning， providing a comparative analysis of their performance. Finally， we highlight significant research achievements of deep learning in the field of video coding， examine the challenges and limitations of existing techniques， and offer an outlook on future trends in video coding technologies.