Car-washing Video Compressed Sensing Technology

With the development of the wireless network technology, the surveillance video enters into more and more industries. In the traditional signal processing, only when the signal samples meet the frequency twice bigger than the highest one can the original signal be reconstructed accurately. This directly causes the problems, such as the big data storage and slow signal transmission. However, Compressed Sensing, a new codec theory brings hopes to the video processing technology. Based on which, the video processing technology has become the research focus home and abroad.In this paper, the car-washing videos would be used as the research background and samples. First, the monitoring videos would be effectively selected by the basic image processing, in order to save the car image which will be further processed with the method of Compressed Sensing. As one of the key technologies in Compressed Sensing, the dictionary construction method affects the quality of signal reconstruction to a large extent. On the basis of the introduction of the framework of Compressed Sensing theory, this paper focuses on the reconstruction algorithm, summarizing the advantages and disadvantages of various reconstruction algorithm. Furthermore, this dissertation analyzes the KSVD dictionary training method in depth, and gives a Compressed Sensing of video processing method combining the KSVD's initial dictionary training method with the OMP algorithm. This method can constantly make atomic iteration and update the dictionary so as to reduce the errors and obtain better reconstruction quality. Considering the correlation between video frames before and after, the paper gives a kind of KSVD dictionary training method which is based on frame differential method. By setting the frame group, the author constantly adjust the sampling rate of key frames and non-critical frames. This would make use of not only the frame information, but the inter frame information efficiently to obtain more significant reconstruction effects of the subjective usual and objective numerial reconstruction.According to the experimental results, the Compressed Sensing approach to car-washing video images enables the storage space to be greatly reduced. Compared with the KSVD dictionary method which does not use the frame difference, the KSVD dictionary on account of the frame difference increases the video image average PSNR (peak signal to noise ration)by 1.86-3.95 dB and improves both the subjective and objective quality of the reconstructed image as well, when the sampling rate is 0.9 and non-critical frame sampling rate is 0.1.