The Design And Implementation Of The Adaptive Distributed Compressed Video Sensing System

Distributed compressed video sensing technology(DCVS),as a novel video coding framework,applies independent encoding at encoders and joint decoding at decoders.It reduces the complexity of encoding by migrating the complex computation from encoders to decoders.Meanwhile,combined with compressed sensing technology,it breaks the limitation of Nyquist sampling rate.Sampling and compression processing can be realized simultaneously.DCVS schemes are widely studied in the research of multimedia communication scenes with resource-constrained encoder.The multihypothesis-based residual reconstruction scheme has attracted widely attention due to its good performance in the research of DCVS.However,current work focus on the theory research with theoretical assumption.The practical application scenarios are less considered.To verify the availability of DCVS,an adaptive DCVS codec system based on practical application scenario is proposed in this thesis.A series of improved schemes are proposed to solve the practical problems in real application.The main work is summarized as follows:Firstly,in current DCVS encoding procedure,the sampling rate of key frames is high,and it reduces the compression efficiency of encoder.Moreover,the correlation between key frame and non-key frame is ignored in DCVS schemes.The redundant information is transmitted to the decoder.To solve the above problems,we divide the image blocks in key frame into three levels according to its texture.Different sampling rates are allocated to each kind of image blocks.The data of key frames to be transmitted is reduced.For non-key frames,a new block encoding mode is proposed to process the blocks with high inter-block correlation.The block has high correlation with its reference block is not encoded and transmitted.It effectively improves the compression efficiency of DCVS scheme and mitigates the transmission pressure.Secondly,this thesis studies the Block Compressed Sensing with Smooth Projected Landweber(BCS-SPL)algorithm.Due to the fixed scale factor applied in threshold-dependent judge procedure,the recovery performance of BCS-SPL fluctuates for video frames with different texture.Considering the relationship between scale factor and image texture,this thesis proposes an adaptive threshold calculation model.By using an adaptively adjusted approximate optimal scale factor to replace the fixed scale factor in threshold calculation,the reconstruction quality of the BCS-SPL algorithm is effectively improved.Thirdly,according to the above improvement,we realize a real and practical adaptive DCVS codec system.And a GPU based reconstruction scheme is proposed to realize the parallel computing of CS reconstruction.It reduces the time delay of CS recovery and improves the computing efficiency in practical application.At last,simulation results show that the proposed DCVS scheme outperforms the existing scheme.The compression efficiency,recovery quality,and computing efficiency are all improved in our proposal.Based on the newly proposed DCVS scheme,a real DCVS codec system is realized in practical application.It verifies that efficiency and the availability of the proposed new DCVS scheme.