Research On Surveillance Video Coding And Super-resolution Reconstruction

With the rapid growth of video surveillance market and network video monitoring system, compression and post-processing analysis of massive digital surveillance video become more and more important. High efficient video coding technology is the key to solve these problems. However, there appeared many problems in the practical application of video surveillance system. First, when video noise increases, the video coding efficiency considerably decreases. Second, the new generation of video coding standard such as H.264/AVC can obtain higher compression efficiency, but it has a very complex coding control model and is difficult for hardware implementation. Third, as the complexity of the new generation of video coding standard is very high, the post-processing and analysis of video surveillance system become more and more difficult. So this thesis is focus on three problems:the noise robustness of video encoder, the complexity of video encoder in the resource-constrained system, compressed video post-processing and enhancement. The main contributions of this thesis are as follows:First, the basic theoretical knowledge of video coding and hybrid coding framework is deeply introduced. And the key technology of China's AVS-S standard which developed for video surveillance system is described. The dissertation pointed out the insufficiencies and omissions of the current research, and described the scope of this research.Second, a noise robustness video coding method for video surveillance system is proposed. By analyzing the noise impact of the H.264 coding model, the thesis pointed out the video coding efficiency decreases caused by the inter-frame predictive coding accuracy rate has dropped. And this paper presents a pre-classification method based on co-matching criteria and motion vector field spatial and temporal filtering. The method uses the co-matching criteria to judge the current macroblock in order to eliminate the noise impact and use the temporal and spatial filtering of the motion vector field of encoded frames to eliminate the noise motion vectors. Finally, according to the motion information of current macroblock, the method limits the coding mode of current macroblock. The proposed algorithm can improve the noise robustness of the H.264 encoder. Simulations show that this approach can result in a time savings of over 62.86% for several typical surveillance sequences. And it also reduces the average Bjontegaard delta bit rate by about 1.67% and increases the average Bjontegaard delta peak signal-to-noise ratio by about 0.08dB when compared with the algorithm of H.264. Experiments prove that this algorithm improves the coding performance and coding speed.Third, this thesis summarizes researches of fast algorithms of H.264 and finds out that many fast algorithms bring a higher computational complexity for the video encoder. In order to reduce the difficulty of the hardware realization of the new generation hybrid coding framework, this thesis propose a fast method based on statistical judgments for fast elimination of redundant prediction modes. This method use the coded block pattern and inter prediction coding cost as the auxiliary information to determine whether to take the current coding mode. And it also uses the statistical distribution for rapid elimination of the coding modes with low probability. This algorithm has a very low computational complexity and optimized the H.264 encoding process. Simulation results reveal that the proposed algorithm can reduce the encoding time by 12.3% on average with the limited performance loss of about 0.037dB. The memory using of this method is very small. And it provides some novel thoughts for designing the video encoder.Finally, a compressed video super-resolution reconstruction algorithm is presented. This paper summarizes researches on compressed video super-resolution reconstruction and points out that the existing MPEG-based approaches are difficult to apply to the new generation video coding technology. The proposed algorithm is based on the variable-pixel reconstruction algorithm which is more suitable for super-resolution reconstruction of video sequence. The algorithm collects the effective macroblock coding side information on the encoder and retains the macroblock information between adjacent frames. In the decoder side, it uses the macroblock mode and motion vector information of the macroblock coding side information to projected the pixels of the adjacent reconstruction frames onto the super-resolution reconstruction image. The spatial resolution, objective performance, and the decoder reconstruction of super-resolution reconstruction of video sequence experiments proved that this algorithm has a better performance of the objective and subjective performance. And this algorithm provides a novel solution idea for compressed video post-processing and analysis:to save the information of the moving object in encoder side for the post-processing and analysis of decoder side.To sum up, this paper digs into the video coding technology for video surveillance system and proposed a set of surveillance video coding solution. On the encoder side, the solution enhanced the encoder noise robustness and improved the coding efficiency. And on the decoder side, it can effectively solve the issue of compressed video enhancement.