基于压缩感知的交通视频压缩技术研究

发布时间：2019-04-10 18:47

【摘要】：交通视频作为智能交通系统的重要组成部分有着广泛的应用,不断产生的庞大交通视频数据,给交通视频的存储带来不小的挑战,如何对交通视频进行压缩就成为一个重要的研究课题。当前交通视频的压缩仍采用传统的基于统计的编码压缩方式,而未能充分利用交通视频的特点对其进行压缩。交通视频具有背景稳定、敏感区域明确、图像纹理复杂等特点,并且交通视频监控通常安装在户外,交通视频图像会受到户外光照变化、天气变化的影响。如何充分利用交通视频的特点,研究出适合交通视频特点的压缩方法就成为重要的研究课题。压缩感知理论为交通视频的压缩提供了一种有益的思路。由于交通视频具有很大的空间冗余和时间冗余,针对这些特点,采用压缩感知理论就可以有效地对交通视频进行观测压缩。根据以上思路,本文对基于压缩感知的交通视频压缩方法进行了研究,所做的具体工作如下:(1)在理解压缩感知理论和相关定理的基础上,重点研究了基于K-SVD算法的交通图像压缩感知重建,针对K-SVD算法时间复杂度高、重构图像质量一般的缺点,本文提出一种基于小波树变迭代次数K-SVD算法。仿真实验结果表明,基于小波树变迭代次数K-SVD算法与原K-SVD算法相比,PSNR值提高2dB左右,算法运行时间降低了15%左右。(2)交通视频预处理是交通视频压缩编码框架设计的基础。在预处理部分,首先对交通视频进行背景建模,背景提取采用混合高斯模型,与均值法相比,所提取的背景更干净清晰;其次对视频进行背景更新,本文使用一种基于分块分类的背景更新方法,在此背景更新算法中,采用三帧差分法获得差分图像,自适应迭代阈值方法确定分类所需的阈值,并用所提取的背景,得到更新背景。第三,对交通视频进行场景分类,对交通视频场景的昼夜进行判断,并对夜间图像进行增强。最后,为提高视频压缩率及视频质量,本文提出一种变采样率计算模型:根据分块压缩感知理论,采用拟合经验函数的变采样率算法,并在此基础上描述一种适用于图像组(Group of Picture,GOP)的变采样率观测压缩过程。(3)设计出一种基于压缩感知的交通视频编码框架,并仿真验证此框架对视频的压缩性能及可用性。
[Abstract]:As an important part of intelligent transportation system, traffic video has a wide range of applications, and the huge traffic video data is constantly generated, which brings a lot of challenges to the storage of traffic video. How to compress traffic video becomes an important research topic. The current traffic video compression still uses the traditional statistical-based coding compression mode, but fails to make full use of the characteristics of traffic video to compress it. Traffic video has the characteristics of stable background, clear sensitive area, complex image texture and so on. Traffic video surveillance is usually installed outdoors, and traffic video image will be affected by outdoor illumination and weather change. How to make full use of the characteristics of traffic video and study the compression method suitable for the characteristics of traffic video becomes an important research topic. The theory of compression perception provides a useful idea for the compression of traffic video. Because traffic video has a lot of space redundancy and time redundancy, according to these characteristics, the compression perception theory can be used to observe and compress traffic video effectively. According to the above ideas, this paper studies the traffic video compression method based on compression perception. The concrete work is as follows: (1) on the basis of understanding the compression perception theory and related theorems, This paper focuses on the traffic image compression perceptual reconstruction based on K-SVD algorithm. In view of the disadvantages of high time complexity and general image quality of K-SVD algorithm, this paper proposes a K-SVD algorithm based on wavelet tree variation iteration times. The simulation results show that, compared with the original K-SVD algorithm, the PSNR value of the K-SVD algorithm based on wavelet tree variation iteration times is about 2dB higher than that of the original PSNR algorithm. The running time of the algorithm is reduced by about 15%. (2) Traffic video preprocessing is the basis of the design of traffic video compression coding framework. In the pre-processing part, first of all, the traffic video background modeling, background extraction using the mixed Gao Si model, compared with the mean method, the extracted background is cleaner and clearer; Secondly, a background update method based on block classification is used in this paper. In the background updating algorithm, the difference image is obtained by three-frame difference method, and the adaptive iterative threshold method is used to determine the threshold required for classification. The background is updated with the extracted background. Thirdly, it classifies the traffic video scene, judges the day and night of the traffic video scene, and enhances the night image. Finally, in order to improve the video compression rate and video quality, this paper proposes a variable sampling rate calculation model: according to the theory of block compression perception, the variable sampling rate algorithm based on fitting empirical function is adopted. On this basis, a variable sampling rate observation compression process suitable for (Group of Picture,GOP is described. (3) A traffic video coding framework based on compression perception is designed. The video compression performance and availability of this framework are verified by simulation.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U495;TP391.41

【参考文献】