基于压缩感知的监控视频重构  

Reconstruction of surveillance video based on compressed sensing

作　　者：李昂[1] 马强[2] 岑翼刚[1] 赵瑞珍[1] 岑丽辉[3] 

机构地区：[1]北京交通大学信息科学研究所,北京100044 [2]宜昌供电公司电能计量中心,湖北宜昌443000 [3]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《智能系统学报》2013年第6期512-516,共5页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(61272028;61104078;61073079);中央高校基本科研业务费资助项目(2013JBZ003);上海交通大学系统控制与信息处理教育部重点实验室开放基金资助项目(SCIP2011009);高等学校博士学科点专项科研基金资助项目(20110162120045;20120009110008);教育部新世纪优秀人才支持计划资助项目(NCET-12-0768)

摘　　要：传统对视频先采样再压缩的方法极大地浪费了硬件资源,针对这一问题,基于压缩感知理论提出了一种对监控视频采样及重构的方法.该方法先获得帧间差值,再将差值投影到小波稀疏域后进行压缩采样.选取每一分组中的中间帧作为关键帧,对关键帧不进行处理,完全保留所有采样点.恢复时利用关键帧和差值可以得到初步重构的视频序列,最后通过运动估计和运动补偿得到优化.实验结果表明,与仅使用压缩感知对差分帧进行重构的方法相比,该方法对监控视频重构帧序列图像的平均峰值信噪比有较大的提升,且受采样点数影响较小,具有很好的鲁棒性.The traditional first-compression-then-sampling method used for video greatly wastes hardware resources. In order to solve this problem, a sampling and reconstruction method used for the surveillance video is proposed based on compressed sensing. With this method, first, the interframe difference is calculated and projected to the sparse domain of the wavelet ,then the wavelet coefficients are sensed by the sensing matrix.The middle frame in every group is taken as the key frame, which is not processed, and all sampling points are completely preserved. In the stage of recovery, the preliminary reconstruction of the video sequence can be obtained by using the key frames and interframe difference. Finally, the optimization is realized by motion estimation and motion compensation. The experimental results show that, compared with the reconstruction for difference frames by only using compressed sensing, this method can greatly lift the mean peak signal-noise ratio of the sequence image on the reconstruction frames of a surveillance video. In addition, the influence caused by the quantity of the sampling points is small, and the robustness is excellent.

关 键 词：压缩感知 监控视频 关键帧 帧间差值 运动补偿 

分 类 号：TN911.73[电子电信—通信与信息系统]