海底光学图像拼接方法研究

发布时间：2018-05-27 16:13

本文选题：海底光学图像 + 图像配准　；参考：《大连海事大学》2017年硕士论文

【摘要】：随着海上运输业、滨海旅游业以及养殖捕捞业的日益活跃和快速发展,船舶、飞机和人员在海上活动的数量、密度不断加大。且近年来,极端天气事件频发,通航环境日益复杂,造成人员伤亡,财产损失的概率随之大大增加。利用AUV携带照相系统对海底进行探测,实现对海底清晰光学成像,随着AUV系统的移动,拍摄大量海底图像序列,而单幅海底图像仅仅能获取和记录有限距离和视角内的信息。通过研究海底光学图像拼接方法,将获取的图像数据拼接形成大范围的水下整体图像,为认知海底整体情形提供帮助,同时为确认水下遇险目标提供技术支撑。本文主要研究海底光学图像拼接方法。考虑到多张图像配准时,选取某张图像作为参考平面时不是最小误差平面,本文提出一种基于最小误差平面的图像配准方法。在图像融合方面,对比几种常用的图像融合方法,选取一种融合效果较好的方法。最后通过实验验证算法的正确性。图像配准是图像拼接的关键环节,传统图像配准方法都是在二维图像平面特征点的基础上进行。当拍摄的图像数量较大,用待配准图像中的某一幅图像作为参考图像并不能解决全局配准问题。针对以上存在问题,本文在传统图像配准方法基础上进行改进,提出一种基于空间点云的图像拼接方法,以空间点云为基础拟合一个最佳投影平面作为参考图像,从而实现水下图像拼接。图像融合技术也是实现图像拼接不可或缺的关键一步。图像融合方法有很多,资料显示在众多的图像融合方法中,基于拉普拉斯金字塔图像融合方法是目前为止融合效果较好的图像融合方法。本文采用拉普拉斯金字塔图像融合方法对配准图像进行融合处理,得到更加自然,更适合人眼视觉的全景图像。最后,为了验证本文算法的适用性和有效性,分别在实验室水池和大连二砣子岛周边海域进行模拟实验,对本文的算法进行验证。
[Abstract]:With the increasingly active and rapid development of marine transportation, coastal tourism and aquaculture and fishing, the number and density of ships, aircraft and personnel at sea are increasing. In recent years, extreme weather events occur frequently, the navigable environment is increasingly complex, resulting in casualties, and the probability of property loss increases greatly. The AUV camera system is used to detect the seafloor and realize the clear optical imaging of the seabed. With the movement of the AUV system, a large number of submarine images are photographed, while a single submarine image can only obtain and record the information in the limited distance and angle of view. By studying the method of submarine optical image stitching, the acquired image data can be stitched into a large range of underwater whole images, which can help to recognize the whole underwater situation and provide technical support for the identification of underwater distress targets. This paper mainly studies the method of submarine optical image stitching. Considering that more than one image is on time and a certain image is not the minimum error plane when a certain image is selected as the reference plane, this paper presents an image registration method based on the minimum error plane. In the aspect of image fusion, compare several common image fusion methods, select a better fusion method. Finally, the correctness of the algorithm is verified by experiments. Image registration is a key link in image mosaic. Traditional image registration methods are based on two-dimensional image plane feature points. When the number of images taken is large, the problem of global registration can not be solved by using one of the images to be registered as a reference image. In view of the above problems, this paper improves on the traditional image registration method, and proposes an image mosaic method based on spatial point cloud, which uses spatial point cloud as the basis to fit an optimal projection plane as a reference image. In order to achieve underwater image stitching. Image fusion technology is also an indispensable key step to achieve image stitching. There are many image fusion methods, and data are shown in many image fusion methods. Image fusion based on Laplacian pyramid is a better image fusion method so far. In this paper, the image fusion method of Laplacian pyramid is used to fuse the registered image to obtain a more natural and more suitable panoramic image for human vision. Finally, in order to verify the applicability and effectiveness of the proposed algorithm, simulation experiments were carried out in the sea area around the laboratory pool and the Dalian Erduzi Island, respectively, and the algorithm was validated.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】