视频配准算法研究

 2022-02-20 07:02

论文总字数:27990字

摘 要

有别于传统的图像配准算法,视频帧间图像配准算法在解决生成大视场全景图问题时,具有高精确性、高实时性等优势。随着一种新型的对地观测手段,即无人机遥感系统,在国土监测、资源勘探、灾害勘察等实际领域的广泛应用,研究一种能在处理海量的视频帧图像的同时兼顾拼接效果与拼接效率的视频配准算法,具有十分重要的现实意义。

在处理视频图像拼接中配准误差的积累问题时,本文使用了一种改进的视频全景图拼接算法,在保证速度和拼接质量的同时提高了信息完整度与精确度,最终开发了一种基于Speeded Up Robust Features的视频配准软件。

首先,对预处理后的原始视频帧间图像进行SURF特征提取,与以往提出的同类算法相比,SURF算法的独特性、鲁棒性与重复度都有着良好的表现,并且其提高运算速度有着显著的提高。

然后,在匹配图像特征时采用最近邻向量匹配法,再利用RANSAC的方式剔除误匹配并算出变换矩阵,利用该矩阵配合双线性插值完成坐标映射,并采用加权平均融合算法,在提高融合效率的同时得到了平滑过渡的融合效果。

最后,在处理多幅图像同时配准的问题时,使用一种将当前帧图像与当前全景图像直接配准的方法,有效地抑制了误差累积,可快速、精确地完成全局配准。

在实现配准的基础上,进一步采用C 编程设计开发了一套基于SURF的视频配准软件。该软件功能为将提供的视频序列的帧图像互相配准,在此基础上生成被测场景的全景图像。该软件在保证配准及融合过程稳定可靠的基础上,用户友好性与可操作性强,运行时间短暂,基本可以满足视频配准要求。

关键词:视频配准;SURF;特征提取;RANSAC;全局拼接误差积累

Research on VIDEO Registration Algorithm

Abstract

Video registration as one of the key technologies in the field of image processing, which is different from the traditional image registration, has advantages like high accuracy and high real-time performance. With the wide use of a new earth observation way that called the UAV remote sensing system in the fields of land monitoring, resource exploration and disaster survey, the research on how to take both the effect and efficiency of splicing into account when dealing with the large amount of the video frame images, has some very pactical worth.

In this paper, to solve the problem of the deviation accumulation during the video image mosaic process, the video panorama stitching algorithms for large static scene observations was improved and the speed is fast while the quality of stitching is high. Ultimately a video registration software which is based on Speeded Up Robust Features is developed.

First, the video frame image which is preprocessed goes through SURF features extraction. SURF algorithm has better performance in repeatability, uniqueness, robustness than similar proposed method, and its computational efficiency advantage is obvious.

Then we get the preliminary matches of image features by using the nearest neighbor vector matching method. RANSAC method is applied to remove mismatches and calculate the transformation matrix. Complete the coordinate mapping by using the matrix along with bilinear interpolation. The use of weighted average fusion improves the efficiency and smoothes the transition of fusion by the same time.

Finally, registrate the current frame image with panoramic image directly when dealing with the multiple images simultaneously registration problem. As a result, the accumulation of errors is effectively inhibited, which means the method can complete global registration quickly and accurately.

On the basis of registration, a SURF-based video registration software programmed with C is developed. The software can play UAV video and stitch aerial images. It has good operability and interactive performance, which is also stable and reliable.

KEYWORDS: Video registration, SURF, feature extraction, RANSAC, global alignment errors accumulate

目录

摘要 I

Abstract II

第1章 绪论 1

1.1 选题意义 1

1.2 视频配准算法概述 1

1.2.1 定义和目的 1

1.2.2 方法分类 2

1.2.3 研究现状 2

1.3 论文的组织安排 3

第2章 SURF特征描述子 5

2.1 尺度空间 5

2.1.1 积分图像 5

2.1.2 HESSIAN矩阵 6

2.1.3 构成三维尺度空间 7

2.2 特征定位 8

2.3 特征主方向 9

2.4 特征描述子计算 9

2.5 实验结果与分析 10

2.6 本章小结 13

第3章 基于特征的视频配准 14

3.1 特征匹配 14

3.2 匹配准则 15

3.3 配准过程及改进 16

3.3.1 RANSAC算法剔除误配点对 16

3.3.2 改进的配准方法 17

3.4 实验结果与分析 18

3.5 本章小结 19

第4章 图像融合 21

4.1 融合基本概念 21

4.2 融合常用算法 21

4.3 加权平均融合 22

4.4 多分辨塔式融合 23

4.5 小波变换融合 23

4.6 实验结果与分析 24

4.7 本章小结 26

第5章 视频配准软件 27

5.1 软件概述 27

5.2 设计与实现 28

5.2.1 图像预处理模块 28

5.2.2 图像拼接模块 28

5.2.3 用户交互模块 29

5.3 测试结果 30

5.4 本章小结 33

第6章 总结与展望 34

6.1 本文主要工作 34

6.2 未来工作展望 34

参考文献 35

致谢 37

绪论

选题意义

视频是一种场景信息的丰富载体,近年来,随着图像处理领域的飞速发展,提出了许多从视频中提取并处理图像信息的方法。与从单张图片提取并处理信息的静态图像配准不同的是,视频序列图像配准能得到形状更加完整、信息更加丰富的目标区域大视场全景图。在仿真虚拟现实、视频内容检索以及获取高分辨率图像等研究领域,视频配准算法有着十分重要的科研意义。

剩余内容已隐藏,请支付后下载全文,论文总字数:27990字

您需要先支付 80元 才能查看全部内容!立即支付

该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;