基于三目的实时视频拼接分析-电路与系统专业论文.docxVIP

摘 摘 要 i i 视频拼接是将若干个摄像头采集的具有重叠区域的视频进行拼接，最终形成 全景视频的技术，相比单目视频，它能提供更大尺寸的视频画面，从而可以同时 展现更多的内容，也为相机阵列的高清显示奠定了基础。随着视频采集设备的不 断廉价化，视频拼接技术成为计算机视觉图像领域的热点研究问题。这项技术在 视频监控、超大图像采样和全景图合成等方面有着非常广泛的应用。然而由于图 像配准和视频码流的复杂性，使得实时视频的拼接成为一个难点问题。本文利用 三个普通摄像头，基于 SIFT 特征提取算法，对实时三目视频拼接技术进行了研 究。 视频流由帧图像组成，因此视频拼接是以图像拼接为基础。但是相比图像拼 接技术视频拼接不仅对拼接的准确性有要求，同时对拼接的实时性也有着较高的 要求。所以本文从图像视频拼接过程中用到的相关基础理论出发。介绍了图像拼 接技术，其中主要对图像的特征提取进行了研究，针对特征点的信息含量高、抗 噪声能力强的特点，确定本文选取基于特征点的特征提取算法来完成实时视频图 像的拼接。 由于视频拼接的核心是视频图像的配准和融合，本文针对特征点提取的 Harris 算法、SIFT 算法以及几种基于 SIFT 算法衍生的特征提取算法进行了深入 的对比研究。最后基于 SIFT 算法的稳定性和鲁棒性，采用 SIFT 算法来完成视 频图像的特征提取。在提取图像特征点后，利用 BBF 算法进行特征配准，并结 合 RANSAC 算法剔除误匹配，然后计算图像间的透视变换矩阵。针对视频拼接 的实时性要求，对基于 SIFT 算法的帧图像拼接技术和视频采集进行了改进。通 过固定摄像头位置采集视频，并结合首帧配准获取透视变换矩阵的方法实现了三 目视频的实时拼接。 关键词：实时视频拼接，图像拼接，SIFT，特征提取，图像配准 Abstract Abstract The video mosaic is a technology that stitches the videos, which are captured by some cameras with overlapping perspectives, and it is used to implement a full-view panoramic video. Compared with monocular video, it can provide greater size of the video picture, and show more content at the same time, also lay the foundations for the camera array of high-definition display. As video acquisition device constantly deflationary force, video mosaic technology become the hotspot in the field of computer vision. The video mosaic technology is widely used in video surveillance, large image sampling, panorama synthesis and other areas. Because of the complexity of the image registration and video streams, the real-time video stitching becomes a difficult problem. This paper just uses three common cameras to research on video of the trinocular in real time based on algorithm of SIFT feature extraction. Video streaming is composed of frame, so video mosaic is based on image matching. But compared with image splicing technology, video stitching not only has requirements of the accuracy of the splicing, as well as the patchwork also have higher requirements. So this article sets out video image stitching in rela