SIA OpenIR  > 光电信息技术研究室
基于李群上聚类的形状匹配算法研究
Alternative TitleShape matching algorithm based on clustering on Lie group
王宏韬
Department光电信息技术研究室
Thesis Advisor刘云鹏
Keyword去雾 暗原色先验 形状匹配 李群 聚类
Pages58页
Degree Discipline控制工程
Degree Name专业学位硕士
2021-05-21
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract随着社会的进步和科学技术的快速发展,无人机在多种行业中的应用凸显出来。无人机具有飞行灵活,受环境影响小,体积较小等优点,因此在无人机上进行视觉监控就具有了低成本、灵活性强和便捷的特点,近年来越来越多行业选择在无人机平台进行视觉监控。无人机的视觉导航技术是无人机实现视觉监控的重要环节,视觉导航为无人机的飞行控制提供重要信息。目前技术发展水平,在空气质量良好无雾、目标成像不发生仿射变形和目标无遮挡等条件较为良好的情况下,利用无人机架设图像采集装置实现的目视觉导航算法较为成熟,可以满足一定使用情况。但在现实使用过程中,空气质量差空中有雾和烟、目标成像发生仿射变形和目标被遮挡等情况时有发生,导致无人机的视觉导航算法效果不佳,跟踪难度大幅度增大。根据此情况,本文在获取无人机对地面进行拍摄后的图片,首先对图片进行预处理,然后采用并改进形状匹配算法。尤其针对有烟雾、仿射变形、物体遮挡三种识别过程中较为复杂的情况进行相应算法上的研究。在视觉导航中经常会遇到森林着火等问题,由于森林着火而产生的大量烟雾,直接影响了无人机设备的图像采集,因此针对大气图像去雾技术有着重要的意义。本文提出一种基于非分光红外检测技的大气光值估计方法,首先采集分析烟雾环境下的气体浓度变化特征,根据气体特性选择合适的权重系数估算大气光值,最后通过暗通道先验的数字图像去雾技术得到还原度高,分辨率强的图像。现有的形状匹配算法只考虑了匹配形状间的局部特征,没有利用局部特征之间的几何变换一致性约束,因此易导致出现错误匹配。根据局部特征的几何变换一致性,提出一种基于李群上聚类的形状匹配算法。该算法首先将待匹配形状分解为若干片段,采用动态规划算法对片段序列进行粗匹配,同时获取匹配片段间的仿射变换矩阵。进一步在李群GA(2)上对仿射变换矩阵进行聚类,依据几何变换一致性约束对错误匹配进行剔除,实现形状的精匹配。在标准数据库下的实验结果表明,与未考虑几何变换一致性的算法相比,基于李群上聚类的形状匹配算法可以大幅度提高形状识别率和准确率,并且可以增强匹配算法的鲁棒性。
Other AbstractWith the progress of society and the rapid development of science and technology, the application of UAV in a variety of industries has become prominent. The UAV has the advantages of flexible flight, small environmental impact, small size, etc., so the visual monitoring on the UAV has the characteristics of low cost, strong flexibility and convenience. In recent years, more and more industries choose the UAV platform for visual monitoring. Visual navigation technology of UAV is an important link to realize the visual monitoring of UAV. Visual navigation provides important information for the flight control of UAV. At the current level of technology development, under the conditions of good air quality and no fog, no affine deformation of the target imaging and no occlusion of the target, the visual navigation algorithm realized by setting up the image acquisition device of the UAV is relatively mature, which can meet certain usage conditions. However, in the real use process, the air quality is poor, there is fog and smoke in the air, affine deformation of the target imaging and the target is blocked from time to time, which leads to the poor effect of the visual navigation algorithm of the UAV and greatly increases the difficulty of tracking. According to this situation, in this paper, after obtaining the pictures taken by the UAV on the ground, the image is preprocessed at first, and then the shape matching algorithm is adopted and improved. Especially for the smoke, affine deformation, object occlusion in the three kinds of recognition process of the more complex case on the corresponding algorithm research. In visual navigation, problems such as forest fires are often encountered. A large amount of smoke caused by forest fires directly affects the image acquisition of UAV equipment. Therefore, the technology of atmospheric image defogging is of great significance. This paper proposes a nondispersive infrared detection technology based on the atmospheric light value estimation method, the first collection and analysis of smoke gas concentration change characteristics of environment, according to the gas characteristics to choose the appropriate weight coefficient estimate atmospheric optical value, finally through the dark channel prior reduction degree is high, the digital image to fog technology is strong resolution images. The existing shape matching algorithms only consider the local features of the matching shapes, and do not use the geometric transformation consistency constraint between the local features, so it is easy to cause the wrong matching. According to the consistency of geometric transformation of local features, a shape matching algorithm based on Lie group clustering is proposed. In this algorithm, the shape to be matched is decomposed into several segments based on corner detection, and the dynamic programming algorithm is used to carry out rough matching of the segment sequences. At the same time, the affine transformation matrix between the matched segments is obtained. Furthermore, the affine transformation matrix is clustered on Lie group GA(2), and the error matching is eliminated according to the consistency constraint of geometric transformation, so as to achieve accurate shape matching. Experimental results in a standard database show that, compared with the algorithm without considering the consistency of geometric transformation, the shape matching algorithm based on Lie group clustering can greatly improve the shape recognition rate and accuracy, and can enhance the robustness of the matching algorithm.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/28969
Collection光电信息技术研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
王宏韬. 基于李群上聚类的形状匹配算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2021.
Files in This Item:
File Name/Size DocType Version Access License
基于李群上聚类的形状匹配算法研究.pdf(2533KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王宏韬]'s Articles
Baidu academic
Similar articles in Baidu academic
[王宏韬]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王宏韬]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.