图索引结构词袋模型的无人机影像匹配对检索

doi:10.13474/j.cnki.11-2246.2023.0110

测绘通报 ›› 2023, Vol. 0 ›› Issue (4): 93-98.doi: 10.13474/j.cnki.11-2246.2023.0110

• 无人机测绘技术应用推广 • 上一篇下一篇

图索引结构词袋模型的无人机影像匹配对检索

刘思康¹, 郭丙轩¹, 姜三², 鄢茂胜¹

1. 武汉大学测绘遥感信息工程国家重点实验, 湖北武汉 430079;
2. 中国地质大学(武汉), 湖北武汉 430074

收稿日期:2022-05-11 发布日期:2023-04-25
通讯作者: 郭丙轩。E-mail:mobilemap@163.com
作者简介:刘思康(1997—),男,硕士生,研究方向为实景三维重建。E-mail:2020206190043@whu.edu.cn
基金资助:
国家自然科学基金重大研究计划(B2021061516)

Matching pair retrieval method of UAV images based on the graph structure bag of words model

LIU Sikang¹, GUO Bingxuan¹, JIANG San², YAN Maosheng¹

1. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China;
2. School of Computer Science, China University of Geosciences, Wuhan 430074, China

Received:2022-05-11 Published:2023-04-25

摘要/Abstract

摘要： 无人机影像匹配对选择是提升影像匹配效率和三维重建稳健性的关键技术。针对经典树状索引结构词袋模型存在查找单词效率低、影像相似度计算精度低、时间复杂度高的问题,本文设计了导航小世界(NSW)图索引结构和TF-IDF-Match4算法,并提出了一种基于图索引结构词袋模型(GSBoW)的无人机影像匹配对检索方法。首先,利用SIFT GPU算法提取无人机影像特征,并通过分层K-means进行特征描述子集合聚类生成单词;然后,利用NSW索引结构进行单词组织,即从单词集合中随机挑选单词作为顶点插入图中,同时找到最邻近M个顶点建立顶点之间的边连接关系,直至所有单词插入结束;最后,在GPU端利用NSW索引结构进行最邻近单词检索,并使用TF-IDF-Match4算法计算查询影像与数据集影像的相似度,实现无人机影像的匹配对选择。本文利用3组大规模航空无人机影像进行试验,并与Colmap和DBoW的词袋模型算法进行对比。结果表明,与Colmap和DBoW词袋模型相比,本文的GSBoW检索算法效率分别提高了45和18倍,且显著提高了初始匹配精度。本文方法提供的影像匹配对能够保证三维重建获得更高的精度。

关键词: 影像检索, 词汇树, 导航小世界, TF-IDF-Match4加权, GPU, 最邻近查找

Abstract: Matching pair selection is a key technology to improve feature matching efficiency and 3D reconstruction reliability of UAV images. However, the classical tree index structure word bag model has low efficiency in finding words, low precision in similarity calculation, and high time cost in image retrieval. This study designs the navigation small world (NSW) graph index structure and TF-IDF-Match4 algorithm, and proposes matching pair retrieval method based on the Graph Structure Bag of Words (GSBoW) model. Firstly, the SIFT GPU algorithm is used to extract features of UAV images, which are used to generate visual words through hierarchical K-means clustering. Secondly, visual words are indexed by using a NSW graph index structure, which is achieved by iteratively selecting a random word and inserting it into the NSW graph, and searching its M nearest vertices that are used to build the edge connection. Finally, the NSW graph structure is implemented on GPU for nearest word searching, and match pair selection is achieved by an efficient algorithm, termed TF-IDF-Match4, for the calculation image similarity scores. The experiments are carried out using three large-scale UAV datasets and compared with the bag of words model algorithms in Colmap and DBoW. The results show that the proposed match pair retrieval algorithm can respectively achieve the speed up ratio of 45 and 18 times compared with Colmap and DBoW, which is provided matching pairs for higher accurate 3D reconstruction.

Key words: image retrieval, vocabulary tree, navigation small world, TF-IDF-Match4 weighted, GPU, the nearest neighbor search

中图分类号:

P237

刘思康, 郭丙轩, 姜三, 鄢茂胜. 图索引结构词袋模型的无人机影像匹配对检索[J]. 测绘通报, 2023, 0(4): 93-98.

LIU Sikang, GUO Bingxuan, JIANG San, YAN Maosheng. Matching pair retrieval method of UAV images based on the graph structure bag of words model[J]. Bulletin of Surveying and Mapping, 2023, 0(4): 93-98.

参考文献

[1] 李宇昊. 无人机在林业调查中的应用实验[J]. 林业资源管理, 2007(4):69-73.
[2] 王根铎,韩婷娜,郭国明,等. 无人机航空遥感系统在灾害应急救援中的应用探讨[J]. 科技传播,2014,6(14):197,147.
[3] 曲林,冯洋,支玲美,等. 基于无人机倾斜摄影数据的实景三维建模研究[J]. 测绘与空间地理信息,2015,38(3):38-39.
[4] SJA B,CHENG J C,WJD E. Efficient structure from motion for large-scale UAV images:a review and a comparison of SfM tools[J]. ISPRS Journal of Photogrammetry and Remote Sensing,2020,167:230-251.
[5] 姜三,许志海,张峰,等. 面向无人机倾斜影像的高效SfM重建方案[J]. 武汉大学学报(信息科学版),2019,44(8):1153-1161.
[6] DATTA R,JOSHI D,LI Jia,et al. Image retrieval[J]. ACM Computing Surveys,2008,40(2):1-60.
[7] HALAWANI A,TEYNOR A,SETIA L,et al. Fundamentals and applications of image retrieval:an overview[J]. Datenbank-Spektrum,2006,18:14-23.
[8] NISTÉR D,STEWÉNIUS H. Scalable recognition with a vocabulary tree[J]. Proceedings of 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'06).[S.l.]:IEEE,2006:2161-2168.
[9] LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision,2004,60(2):91-110.
[10] LEPETIE V,LAGGER P,FUA P. Randomized trees for real-time keypoint recognition[C]//Proceedings of 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05). San Diego,CA,USA:IEEE,2005.
[11] OBDRZALEK S,MATAS J. Sub-linear indexing for large scale object recognition[C]//Proceedings of the British Machine Vision Conference 2005. Oxford:British Machine Vision Association,2005.
[12] SIVIC J,ZISSERMAN A. Video Google:a text retrieval approach to object matching in videos[C]//Proceedings of the 9th IEEE International Conference on Computer Vision. Nice,France:IEEE,2008:1470-1477.
[13] BRODER A Z. On the resemblance and containment of documents[C]//Proceedings of Compression and Complexity of Sequences 1997. Salerno:IEEE,2002:21-29.
[14] RAM P,GRAY A G. Maximum Inner-Product Search using Tree Data-structures[EB/OL].[2022-03-29]. https://arxiv.org/abs/1202.6101v1.
[15] FU CONG,XIANG CHAO,WANG CHANGXU,et al. Fast approximate nearest neighbor search with the navigating spreading-out graph[EB/OL].[2022-03-29]. https://arxiv.org/abs/1707.00143.
[16] MALKOV Y A,YASHUNIN D A. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(4):824-836.
[17] MALKOV Y,PONOMARENKO A,LOGVINOV A,et al. Approximate nearest neighbor algorithm based on navigable small world graphs[J]. Information Systems,2014,45:61-68.
[18] PONOMARENKO A,MAL Y,KOV,et al. Approximate nearest neighbor search small world approach[C]//Proceedings of the International Conference on Information and Communication Technologies and Applications. Florida:[s.n.],2011.

图索引结构词袋模型的无人机影像匹配对检索

Matching pair retrieval method of UAV images based on the graph structure bag of words model

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 7

编辑推荐

Metrics

本文评价

[1]	徐云耘, 周琛, 李满春. CPU+GPU异构环境下数据密集型矢量多边形地理大数据并行框架[J]. 测绘通报, 2022, 0(5): 110-119.
[2]	刘振涛, 杨毅, 王东超, 谢晓尧. 一种基于CUDA的大数据量地理加权回归并行加速算法[J]. 测绘通报, 2020, 0(12): 1-5.
[3]	强永刚, 肖志峰, 陈欢欢, 闫丽阳. 基于动态阈值哈希的大规模遥感影像快速内容检索方法[J]. 测绘通报, 2019, 0(8): 34-38,53.
[4]	孙钰珊, 艾海滨, 韩晓霞. “高分一号”卫星影像业务流程化处理平台研究[J]. 测绘通报, 2018, 0(5): 65-70,135.
[5]	胡屹群, 周绍光, 岳顺, 王莎. 利用视觉词袋模型和颜色直方图进行遥感影像检索[J]. 测绘通报, 2017, 0(1): 53-57.
[6]	程博艳刘强李小文张涛. 利用CUDA实现矢量地图栅格化的并行处理[J]. 测绘通报, 2014, 0(11): 97-101.
[7]	刘旭春李德龙刘津刘杨滕长胜靖常峰. 基于可编程GPU的遥感影像快速处理研究[J]. 测绘通报, 2013, 0(6): 22-26.