Research On Real-time Video Compression Of Space CMOS Camera With High Resolution

Due to the continuous improvement in the video resolution and frame rate and other indicators of space camera, leading to a sharp increase in the amount of video data for CMOS output. Existing compression algorithms mainly stay in the software simulation stage, whose computing process is very complex, and the hardware implementation is difficult and less effective. In order to effectively alleviate the pressure on the satellite channel transmission and storage, develop a good performance, fast data processing, and real-time video compression system is imminent.Under the background of a participated project, the research of space CMOS camera technology, panchromatic and multispectral CMOS camera video compression technologies are studied. Now the main contents and results are summarized as follows:1. According to the project requirements and CMOS video features, the ASIC compression technology based on ADV212 is discussed for small CMOS camera and high-resolution CMOS camera. First and foremost, the Custom-specific mode is used for various video formats, considering the CMOS video format and the working principle of the ADV212. Through the FPGA internal RAM and DDR3 SDRAM Ping-Pong operation, the data can be cached. The working efficiency is greatly improved. Secondly, this method can realize both the direct code stream transmission and the transmission after storage. And through the error correcting coding, the correction ability of the flash memory is greatly improved. Lastly, to verify the method, the compression and de-compression circuit board are involved to specify the performance of the method. The results show that the compression system realizes real-time and stable performance, And the arbitrary compression ratio could be realized through setting the program, the average peak signal to noise ratio (PSNR) is higher than 28 dB when the compression ratio is 80:1, the average PSNR is higher than 26 dB when the compression ratio is 150:1. The compression system can be realized and the real-time performance is good enough with large amount of data.2. An effective compression algorithm of multispectral image based on discrete wavelet transform (DWT) and improved Karhunen-Loeve transform (KLT) is proposed in this paper. It could achieve better performance with effective combination of the two methods which convert the energy of images to a small number of coefficients. In order to reduce the most redundancy of space, it deals with each spectrum with fast 5/3 2-D DWT firstly. Then the spectral 1-D KLT is applied on the 2-D DWT coefficients to reduce spectral redundancy and remnant spatial redundancy. Finally, entropy coding is executed to obtain the compressed code stream. Within 0.25-2bit/pixel (b/p) range, the average PSNR of the proposed compression algorithm could improve greatly through comparing with the other approaches, which reduces the operating time and improves the performance of multi-spectral image compression algorithm, hardware implementation strategies is also proposed, verifies the correctness and feasibility of our method. The work in this paper has a good reference value and significance for the similar case.3. After the image KLT and one level DWT, relevance still exists within different coefficients of spectral bands and high-frequency sub-band of the same spectrum, whose average value is greater than 0.9, this value will be decreased through increasing the wavelet transform levels, however, the effect is not very good; compression algorithm based on DWT and Tucker decomposition takes image as a tensor, which can represent a complete high-dimensional data and maintain their intrinsic structure, it is preferable to remove the image of the spatial and spectral redundancy, but sparse representation is insufficient to reach a higher compression ratio. The proposed algorithm could overcome the shortcomings of the above two methods, ensuring higher compression performance, while effectively protecting the spectral information. Firstly, all the multispectral bands are made the transform of KLT for eliminating the spectral redundancy. Then the 2D discrete 9/7 wavelet transform is done to the transformed spectrum to eliminate the spatial redundancy. Thirdly, for further eliminating the spectral redundancy and spatial residual redundancy, we take every transformed wavelet subband as non-negative tensor, and make the Tucker Decomposition (TD) using dGN to find the optimal solution. In the end, we encode the model matrixes and the core tensor with entropy coding. Within the scope of the compression ratio 4:1 to 32:1, the average PSNR is over 43 dB, which greatly improves the system PSNR, and improves the performance of the multispectral image compression algorithm compared with other multispectral image compression algorithms.4. A camera with small-size and low-power is designed, using a CMOS MT9V032 as digital image sensor. Based on the primary aberration theory, the optical system with small size and compact structure is designed whose focal length is 12.95 mm and F number is 5, and the modulation transfer function (MTF) of all fields are better than 0.5 at the spatial frequency of 83 1p/mm; We use FPGA as the timing control platform for the electronic system and digital video signal of CMOS output which is delivered through the differential chip in low voltage differential signal (LVDS) output format to the image acquisition card. Finally, we can see the image on the computer. Experimental results show that the designed CMOS camera has good image quality, low power, portability, and high reliability, the frame rate is 60 frames/s when the clock is 26.6 MHz, and it could achieve a variety of modes by adjusting the value of the internal registers, which is particularly suitable for the high requirements occasion with camera volume and high imaging quality.5. According to the indicators for high-resolution spatial CMOS camera, video compression system based KAC-06040 CMOS camera system is designed for large and medium-sized video compression system, the optical system with small size and compact structure is designed whose focal length is 1175 mm and F number is 6.71, and the MTF of all fields are better than 0.446 with obscuration at the spatial frequency of 106.41 lp/mm, energy concentration is better than 80% for less than 11um energy concentration; We use FPGA as the timing control platform for the electronic system and digital video signal of CMOS output which is delivered through the Camera Link with medium mode to the image acquisition card. Finally, we can see the image on the computer. Experimental results show that the designed CMOS camera has good image quality, low power, portability, and high reliability, which meet the needs of the project.