面向监控视频编码的分类率失真优化方法研究

As the video surveillance is widely applied in the field of production safety, intelligent transportation and safe city, the massive video data have put forward new challenges to the surveillance video coding. As a core technology of the video coding, rate distortion optimization (RDO) plays a crucial role in affecting the performance of video compression. This project studies on RDO of the surveillance video coding, and proposes the theories and methods of a differentiated RDO framework. In view of the rate-distortion performance differences between the foreground and background, a sub-SAD based feature and a minimum error rate based Bayesian decision method are presented for coding block classification. An adaptive RDO mode is proposed for coding the foreground blocks, in which the background blocks are removed for the interference of the mode estimation. For the long-term background blocks, the distortion propagation mode is introduced to construct a global RDO scheme. Compared with the traditional methods, the project is aimed at a differentiated RDO framework for the rate-distortion performance differences between surveillance video blocks. It is expected to provide new idea and theoretical basis for the surveillance video compression.

随着视频监控广泛地应用于安全生产、智慧交通和平安城市等场合，海量的视频数据对监控视频编码提出了新的挑战。率失真优化作为视频编码的核心技术，对于影响视频压缩性能的好坏起到至关重要的作用。本项目针对监控视频编码的率失真优化问题展开研究，提出面向监控视频编码的分类率失真优化理论和方法。针对前景块和背景块之间的率失真特性差异，提出基于子块绝对差值方差的编码块分类特征，构建基于最小错误率的贝叶斯分类检测方法；针对前景块之间的个体差异，提出基于自适应率失真优化模型的前景块编码方法，并通过消除背景块的干扰提高了模型估计的准确性，进而提升前景块的编码性能；通过引入失真传递模型，提出针对监控视频长时背景块的全局率失真优化模型，以提升背景块的编码性能。与传统方法相比，本项目针对监控视频编码块的率失真特性差异进行分类优化，有望为提升监控视频压缩性能提供新的思路和理论依据。

本项目研究了面向监控视频编码的率失真优化编码方法，研究包括：1）研究监控视频背景失真传播特性，提出背景失真传播模型；2）研究监控视频失真传播比衰减模型，提出基于周期性增强帧的全局率失真优化编码策略；3）研究在分层预测编码结构中应用增强帧的局限性，提出基于长短时参考帧的预测编码结构；4）研究基于视频的动态点云压缩中不同类型编码单元的率失真特性差异，提出基于占位图的动态点云压缩编码分类率失真优化方法；5）研究内容自适应变换以及变换量化联合优化策略，提出联合变换量化的视频编码新方法。项目所提方法显著提高了监控视频的编码效率。 相关成果发表论文14篇，其中SCI检索论文8篇，EI检索国际会议论文6篇，申请国家发明专利8件，其中授权专利3件；协助培养硕士研究生5名。