Research On Distributed Video Compressive Sensing Reconstruction Based On Spatial And Temporal Characteristics

In the21st century, with the rapid development of information technology, various videoapplications are becoming the main service in the Internet and the wireless multimediatransmission. Even higher requirements put forward for the speed and quality of information,which means that the system must have greater power for the signal acquisition and processing.Traditional pattern of data acquisition and compression requires large storage space and has veryhigh calculation complexity. The emerging theory of compressive sensing makes full use of thesparse characteristic of the signal, compresses the signal at the same time while it is sampling,thus effectively reducing the storage requirements and signal processing time, provides a newway for the massive data sampling and processing.Researchers has made a lot of research on the selection of sparse basis, the construction ofmeasurement matrix and the design of reconstruction algorithms during the existing results ofvideo compressive sensing, but as to improving the efficiency of signal acquisition, especiallyreducing the storage requirements of the related measurement matrix and simplifying thecalculation, it still needs further exploration. Besides, the accuracy and the convengence speed ofthe reconstruction algorithms should be optimized with more works.The main contents of the study are summarized as follows:Firstly, an adaptive measurement rate setting strategy is proposed to classify all blocks in avideo frame into static blocks and motion blocks depending on the temporal correlation of videos,and then to set the different measurement rates for them denpending on the block classification,this program can measure video blocks with complex characteristics, thus avoiding uselessmeasurements, enhancing the efficiency of capturing information.Sencondly, the predictive model realizes the motion estimation in the measurement domain,and employs the motion information to construct the sparse representation dictionary of theblocks to be reconstructed, atoms in the dictionary are highly relevant to these blocks, so theperformance of the CS reconstruction algorithm can be improved effectively by using them formultiple hypotheses.Finally, simulation experiments with Matlab are carried out. The results show that theproposed algorithm has a promotion in the aspect of PSNR, SSIM and time for reconstructionunder various video test sequences, which means that the algorithm could, to some extent, optimize the accuracy of the reconstruction algorithms and improve efficiency of thereconstruction.