基于场景流形的长视频分段拼接的研究与实现-信号与信息处理专业论文.docxVIP

摘要 摘要 I万方数据 I 万方数据 摘要 随着电子信息产业的发展和技术的进步，全景图拼接技术已成为计算机视觉和虚 拟现实中的一种重要的场景表示方法，它广泛应用于航空和遥感视频图像处理、虚拟 现实、视频压缩、视频检索、医疗图像处理等领域。 本文主要研究含有运动物体的长视频全景图的拼接。对现有算法进行分析和改 进，具体研究内容如下： 1、研究了基于场景流形算法的视频全景图拼接技术。针对场景流形算法采用 Dijkstra 寻找最小变形的流形时比较耗时，本文对 Dijkstra 算法在时间和空间效率方 面进行改进，在一定程度上提高了时间效率，节约了内存空间。 2、针对采用场景流形算法进行长视频序列的拼接时内存需求较大和时间效率不 高的问题，本文提出了分段的思想。将长视频序列分段，分段后对每段视频分别采用 场景流形算法并行拼接。试验结果表明，将长视频序列分段可以在保证拼接效果的前 提下，加快拼接速度。 3、每段视频采用场景流形算法拼接后，需要将所有拼接得到的图像缝合成一幅 全景图。在图像融合时，为了避免在运动物体区域出现鬼影，本文研究了最佳缝合线 的图像融合算法：动态规划的最佳缝合线搜索算法和图切割的最佳缝合线搜索算法。 通过分析比较，本文采用图切割算法进行最佳缝合线搜索。 本文在 VC++和 MATLAB 编程环境下实现了长视频序列的拼接。通过实验验证 长视频分段可以在很大程度上提高时间效率，适用于较长视频的全景图拼接。 关键词：全景图拼接；长视频分段；场景流形算法；图切割算法；图像融合. Abst Abstract II万方数据 II 万方数据 Abstract As the rapid development of information technology and electronics, panoramic image mosaic has become an important way to study virtual reality and computer vision. It has wide applications in the fields of aviation and remote sensing image processing, virtual reality, video compression, video indexing, medical image processing and so on. This thesis concentrates on long video image mosaic with moving objects. Analyzing and improving the existing algorithms,the main research points in the paper are as follows: 1、 The thesis studies video image mosaic based on scene manifolds algorithm. However, it is more time consuming when adopting Dijkstra algorithm in scene manifolds algorithm to find the manifold that will incure the fewest distortion. The paper makes the improvement of the Dijkstra algorithm in time and spatial aspect, which in some degree can enhance the time efficiency and save the memory space. 2、 Aimed at the problems of high memory requirement and time consuming when scene manifolds algorithm is purposed to make long video sequences mosaic, the thesis proposes the idea of segemtation. The long video sequence is devided into several parts, and then the method of scene manifolds algorithm is adopted for each section in parallel. The experiment results demonstrate that the method