计算资源受限条件下的监控视频编码与重建方法研究

High definition intelligence video surveillance defined the important future direction of the security monitoring field, and its low complexity coding and super resolution reconstruction are the hot spot of the problem to be solved. The computing resource of front monitor is very limited, which often led to the huge cutting out of the original coding standard features. And the illumination condition of front monitor is also very bad and complex. All of that will lead to the distortion of the reconstruction video quality and can not satisfy the needs of security business. This project will analyze the coding computing resources consumption problem and decoding reconstruction problem as a whole, and will proposal the video coding method based on statistical properties and the super-resolution reconstruction method based on sparse coding theory. The main research contents and innovative points include: 1) we will build the coding mode cost measurement model of the basic coding unit in the encoder, and establish the mixed probability distribution judgment criterion to optimize the whole encoding process with extremely low computational cost and reduce resource needs as well as ensure video quality of reconstruction; 2) we will break the traditional special framework of super-resolution reconstruction in decoder, and use the sparse expression model to directly compensate the high frequency coefficients which are losing by quantization in the transform domain. That will obtain the high resolution image finally. This project strive for the original research achievements, and will effectively reslove the contradiction between the computing resource and the video reconstrction quality. This project will improve the basic research level of china security monitoring field.

高清智能视频监控是未来安防领域的重要发展方向，其低复杂度编码与高清晰度重建是有待解决的热点问题。监控前端的计算资源受到严格限制，往往采用对原编码标准特性大量裁剪的方法控制计算复杂度，加上监控光照条件复杂恶劣，导致重建视频质量下降，无法满足安防业务需求。本项目将编码计算资源消耗和解码重建质量作为整体研究，提出基于编码统计特性与稀疏表示理论的高效视频编码与超分辨率重建方法，主要研究内容与创新之处包括：1)在编码端通过构建基本编码单元的编码模式开销观测模型，建立混合概率分布判决准则，以极低计算代价优化整体编码流程，降低资源需求，保证重建视频质量；2)在解码端突破传统超分辨率空域重建框架，引入稀疏表达模型，直接在变换域中对因变换量化损失的高频系数进行补偿，最终解算出高分辨率图像。本项目力争取得具有原创性的研究成果，有效解决计算资源开销与视频重建质量之间的矛盾，提高我国安防领域的基础研究水平。

本项目针对计算资源受限条件下的视频通信系统在安防监控、应急指挥等国民重点行业中的应用问题，将编码计算资源消耗和解码重建质量作为整体研究，系统梳理了视频通信系统的质量要素，提出了一套三层次视频通信系统音视频质量评价体系，从而解决了以往评估系统无法清晰地分辨出视频通信系统各个环节质量要素的问题。在此基础上，构建了一套完整的视频通信系统编码端和解码端重构与分析实验平台。利用该平台，项目组开展了低计算复杂度开销下视频编码器系统优化、视频编解码器的噪声鲁棒性和视频通信后处理增强重建三方面的研究。针对计算资源和编码质量之间的矛盾，提出了一种利用大规模MIMO系统的多节点无线视频通信高效传输构架和一种基于不等差保护措施的调制优化传输构架，可在相同重建视频质量的条件下，降低编码发送端的能量消耗和提高视频流传输效率；针对视频监控复杂采集场景带来的编码效率降低问题，提出了一种噪声鲁棒的视频编码优化算法，在提高编码速度的同时降低了编码码流；针对安防监控系统后处理视频增强的实际需求，提出了一种基于图像分块纹理信息的双树离散小波硬阈值滤波和全变分滤波的自适应加权滤波模型，该方案可显著提升重建图像的主客观质量，同时可有效保留图像的纹理细节；提出了一种基于变换域系数随机置换的压缩采样表示方法，通过变换域系数在各频带内的随机置换，有效地实现了各测量块的稀疏性的均衡。这些方法为视频通信系统的整体质量优化、降低编码端计算资源消耗，提升解码端视频重建质量等方面的应用提供了有力支持。项目组目前已发表论文十余篇，申请国家专利5项，其中2项获得授权。同时项目组积极推进研究成果的应用转化，获得电力行业信息化优秀成果奖一等奖一项。