洗车监控视频压缩感知技术研究

发布时间：2018-09-07 12:20

【摘要】：随着无线网络技术的发展,监控视频出现在了越来越多的行业领域,传统的信号处理中,信号采样要求满足大于信号最高频率的两倍,这样才能够精准地重构原始信号,这直接导致了数据存储量大,传输信号慢等问题。而压缩感知理论的提出给视频处理技术带来了新的进展,基于压缩感知方法的视频处理技术,逐步成为了海内外的研究热门。本文以洗车行监控视频作为研究背景和样本,首先通过图像处理基本手段将监控视频内容进行了有效筛选,将有车图像保存,对保存图像采用压缩感知的方法处理。字典构造方法是压缩感知十分重要的技术手段之一,对重构信号的质量有着重要的影响。本文一开始先介绍了压缩感知基本理论的框架,对重建算法予以重点介绍,总结各种重建算法的优缺点。对KSVD字典训练算法进行深入分析,并给出了一种结合KSVD初始字典训练法和OMP算法的压缩感知视频处理方法,该方法与能够让原子不断迭代,不断更新字典,达到减小误差,获得更好的重建质量;考虑到视频前后帧间关联性,接下来给出一种基于帧差法的KSVD字典训练构造方法,并利用设置帧组,不断调整关键帧以及非关键帧的采样率,不仅利用了帧内信息,还高效的利用到帧间信息,大大减小了存储空间,并获得了更为显著的主观视觉重建效果和客观数值对比的重建效果。实验结果表明,采用压缩感知方法处理洗车行监控视频图像能够使得存储空间大大减小。并且与未利用帧差法的KSVD字典法相比,在关键帧采样率为0.9时,非关键帧帧差采样率为0.1时,基于帧差法的KSVD字典构造方法使得视频单帧图像的平均PSNR (峰值信噪比)提高了 1.86～3.95dB,提高了重建图像的主观和客观质量。
[Abstract]:With the development of wireless network technology, surveillance video has appeared in more and more industries. In traditional signal processing, the requirement of signal sampling is twice as high as the highest frequency of the signal, so that the original signal can be reconstructed accurately. This directly leads to the problems of large data storage and slow transmission signal. The video processing technology based on compressed sensing has gradually become a hot research topic at home and abroad. This paper takes the monitoring video of the car washer as the research background and sample. Firstly, the content of the monitoring video is effectively screened by the basic means of image processing, and the vehicle image is saved. Preserved images are processed by compressive sensing. Dictionary construction is one of the most important technical means of compressive sensing, which has an important impact on the quality of reconstructed signals. Dictionary training algorithm is analyzed in depth, and a compression sensing video processing method combining KSVD initial dictionary training method and OMP algorithm is proposed. This method can make the atoms iterate and update the dictionary continuously, so as to reduce the error and obtain better reconstruction quality. The KSVD dictionary is trained and constructed by frame difference method, and the sampling rate of key frame and non-key frame is adjusted continuously by setting frame groups. Not only the intra-frame information is utilized, but also the inter-frame information is used efficiently, which greatly reduces the memory space and obtains more significant subjective visual reconstruction effect and the reconstruction effect of objective numerical comparison. The experimental results show that the compression sensing method can greatly reduce the storage space of the video image of the car wash line. Compared with the KSVD dictionary method without frame difference method, when the key frame sampling rate is 0.9 and the non-key frame difference sampling rate is 0.1, the KSVD dictionary construction method based on frame difference method can make the video single frame image level. Average PSNR (peak signal-to-noise ratio) increased by 1.86 to 3.95dB, which improved subjective and objective quality of reconstructed images.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】