Motion Vector Prediction Based Distributed Compressed Video Sensing

In recent years,with the increasing demand for information,the distance between human beings and the physical world has been further narrowed.Wireless multimedia sensor networks have emerged as time requires,and they play a huge role in security monitoring,medical care and other fields.However,the traditional video coding framework is no longer applicable due to the limited computing power at the coding end.The simple encoding and complex decoding of Distributed Video Compressed Sensing has attracted scholar's attention.In the framework of multi-hypothesis prediction technology,the quality of hypothesis sets is particularly important.However,as the motion speed for video sequencing of each block in the same frame is normally different,it is difficult to achieve the reconstruction speed and quality if one was to only use the same hypothesis set acquisition method.In order to solve these problems,a new video coding and decoding framework based on motion vector prediction is proposed in this paper.This scheme can make full use of the correlation in time domain in video sequence,search high-quality hypothesis set quicker,and be able to achieve high-quality reconstruction of video sequence.Firstly,because the motion vectors of adjacent video frames are correlative,therefore when searching for a hypothesis,we would first predict the location of the best similar blocks in the reference frame,and then use this as the center of the search window to search for high-quality hypothesis in the search window.For the motion block,if there is an error in predicting the motion vectors,it will affect the next prediction.In order to avoid erroneous prediction of moving vectors,a search window is reconstructed centering on the corresponding position of the reconstructed block in the reference frame.For stationary blocks,only the search window centered on the best predicted similar blocks is established.At the same time,a new hypothesis set filtering technology is proposed to generate high-quality edge information to assist non-key frames in high-quality reconstruction.Considering the different motion characteristics of intra-frame blocks,if we only use key frames or non-key frames as reference frames may result in poor decoding quality.For stationary blocks,using the key frame as reference frame can provide better quality hypothesis.For moving blocks,adjacent frames in time domain should be used as reference frame to search for high quality hypothesis.Therefore,this paper proposes an adaptive CS frame reconstruction algorithm,which only reconstructs multiple reference frames of moving block,so as to improve the reconstruction quality efficiently.The simulation results show that the proposed distributed video compressed sensing technology based on motion vector prediction improves the reconstruction quality and speed compared with the existing scheme,which proves the feasibility and effectiveness of the proposed scheme.