嵌入式视频流处系统的设计与开发.pdfVIP

万方数据 摘 要 摘 要 近年来，对无人机视频图像进行连续拼接以获得超高分辨率、大视场的全景图在 军事侦察、灾情监测、定位导航等方面有着重要的应用。本文对港口监视目的的无人 机视频图像拼接技术进行研究，设计并实现了基于ARM 的嵌入式视频流处理系统， 该系统自动解码视频流并对关键帧进行拼接，最后合成一幅港口全景图。本文主要内 容如下： 首先，本文在分析传统光流法的局限性基础上，提出了一种融合傅里叶-梅林频 域配准和光流法空域配准的图像配准算法和一种分块光流计算与评估的策略，提高了 光流法的应用范围，解决了摄像机位移幅度较大且图像存在干扰区域时的全局运动估 计问题。 其次，在介绍并分析了视频图像序列拼接的累积误差以及不同拼接合成方式的优 缺点后，本文提出了一种基于金字塔序列的视频图像合成方式，该方式既消除了累积 误差又减小了图像失真程度。 最后，在ARM嵌入式平台下实现了视频流处理系统，该系统自动解码视频流并 利用本文提出的视频图像拼接算法对关键帧进行拼接，最后合成一幅全景图保存并显 示。 关键词：航拍视频拼接 光流法 傅里叶-梅林变换 累积误差 ARM - I - 万方数据 Abstract Abstract In the recent years, the mosaic of UAV video image into an ultra high-resolution panorama with large field of view has been increasingly applied in military reconnaissance, disaster monitoring, navigation and positioning and other fields. This paper makes research on the UAV video image mosaic with the aim of harbor monitoring and has designed and implemented an ARM-based embedded video stream processing system, which can automatically encode the video stream, stitch the key frames and mosaic into the panorama of the harbor. The main contents of this paper are as follows: First, based on the analysis of the limitation of traditional optical flow method, this paper presents an image registration algorithm which combines the Fourier-Mellin registration in frequency domain with the optical flow registration in space domain, and the strategy of patch based optical flow method and assessment. The applic