Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization

Rui Cao; Jiasong Zhu; Qing Li; Qian Zhang; Qingquan Li; Bozhi Liu; Guoping Qiu

doi:10.1007/978-3-030-34120-6_5

Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization

Rui Cao, Jiasong Zhu, Qing Li, Qian Zhang, Qingquan Li, Bozhi Liu, Guoping Qiu

School of Computer Science

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

2 Citations (Scopus)

Abstract

Image-based geolocalization is an important alternative to GPS-based localization in GPS-denied situations. Among them, ground-to-aerial geolocalization is particularly promising but also difficult due to drastic viewpoint and appearance differences between ground and aerial images. In this paper, we propose a novel spatial-aware Siamese-like network to address the issue by exploiting the spatial transformer layer to effectively alleviate the large view variation and learn location discriminative embeddings from the cross-view images. Furthermore, we propose to combine the triplet ranking loss with a simple and effective location identity loss to further enhance the performances. We test our method on a publicly available dataset and the results show that the proposed method outperforms state-of-the-art by a large margin.

Original language	English
Title of host publication	Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1
Editors	Yao Zhao, Chunyu Lin, Nick Barnes, Baoquan Chen, Rüdiger Westermann, Xiangwei Kong
Publisher	Springer
Pages	57-67
Number of pages	11
ISBN (Print)	9783030341190
DOIs	https://doi.org/10.1007/978-3-030-34120-6_5
Publication status	Published - 2019
Event	10th International Conference on Image and Graphics, ICIG 2019 - Beijing, China Duration: 23 Aug 2019 → 25 Aug 2019

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11901 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	10th International Conference on Image and Graphics, ICIG 2019
Country/Territory	China
City	Beijing
Period	23/08/19 → 25/08/19

Keywords

Cross-view geolocalization
Deep metric learning
Image retrieval
Image-based localization

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-030-34120-6_5

Cite this

Cao, R., Zhu, J., Li, Q., Zhang, Q., Li, Q., Liu, B., & Qiu, G. (2019). Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization. In Y. Zhao, C. Lin, N. Barnes, B. Chen, R. Westermann, & X. Kong (Eds.), Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1 (pp. 57-67). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11901 LNCS). Springer. https://doi.org/10.1007/978-3-030-34120-6_5

Cao, Rui ; Zhu, Jiasong ; Li, Qing et al. / Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization. Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1. editor / Yao Zhao ; Chunyu Lin ; Nick Barnes ; Baoquan Chen ; Rüdiger Westermann ; Xiangwei Kong. Springer, 2019. pp. 57-67 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{63d101a09ba147c8abc6b7ce8fcbb3ad,

title = "Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization",

abstract = "Image-based geolocalization is an important alternative to GPS-based localization in GPS-denied situations. Among them, ground-to-aerial geolocalization is particularly promising but also difficult due to drastic viewpoint and appearance differences between ground and aerial images. In this paper, we propose a novel spatial-aware Siamese-like network to address the issue by exploiting the spatial transformer layer to effectively alleviate the large view variation and learn location discriminative embeddings from the cross-view images. Furthermore, we propose to combine the triplet ranking loss with a simple and effective location identity loss to further enhance the performances. We test our method on a publicly available dataset and the results show that the proposed method outperforms state-of-the-art by a large margin.",

keywords = "Cross-view geolocalization, Deep metric learning, Image retrieval, Image-based localization",

author = "Rui Cao and Jiasong Zhu and Qing Li and Qian Zhang and Qingquan Li and Bozhi Liu and Guoping Qiu",

note = "Publisher Copyright: {\textcopyright} 2019, Springer Nature Switzerland AG.; 10th International Conference on Image and Graphics, ICIG 2019 ; Conference date: 23-08-2019 Through 25-08-2019",

year = "2019",

doi = "10.1007/978-3-030-34120-6_5",

language = "English",

isbn = "9783030341190",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "57--67",

editor = "Yao Zhao and Chunyu Lin and Nick Barnes and Baoquan Chen and R{\"u}diger Westermann and Xiangwei Kong",

booktitle = "Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1",

}

Cao, R, Zhu, J, Li, Q, Zhang, Q, Li, Q, Liu, B & Qiu, G 2019, Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization. in Y Zhao, C Lin, N Barnes, B Chen, R Westermann & X Kong (eds), Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11901 LNCS, Springer, pp. 57-67, 10th International Conference on Image and Graphics, ICIG 2019, Beijing, China, 23/08/19. https://doi.org/10.1007/978-3-030-34120-6_5

Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization. / Cao, Rui; Zhu, Jiasong; Li, Qing et al.
Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1. ed. / Yao Zhao; Chunyu Lin; Nick Barnes; Baoquan Chen; Rüdiger Westermann; Xiangwei Kong. Springer, 2019. p. 57-67 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11901 LNCS).

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization

AU - Cao, Rui

AU - Zhu, Jiasong

AU - Li, Qing

AU - Zhang, Qian

AU - Li, Qingquan

AU - Liu, Bozhi

AU - Qiu, Guoping

PY - 2019

Y1 - 2019

N2 - Image-based geolocalization is an important alternative to GPS-based localization in GPS-denied situations. Among them, ground-to-aerial geolocalization is particularly promising but also difficult due to drastic viewpoint and appearance differences between ground and aerial images. In this paper, we propose a novel spatial-aware Siamese-like network to address the issue by exploiting the spatial transformer layer to effectively alleviate the large view variation and learn location discriminative embeddings from the cross-view images. Furthermore, we propose to combine the triplet ranking loss with a simple and effective location identity loss to further enhance the performances. We test our method on a publicly available dataset and the results show that the proposed method outperforms state-of-the-art by a large margin.

AB - Image-based geolocalization is an important alternative to GPS-based localization in GPS-denied situations. Among them, ground-to-aerial geolocalization is particularly promising but also difficult due to drastic viewpoint and appearance differences between ground and aerial images. In this paper, we propose a novel spatial-aware Siamese-like network to address the issue by exploiting the spatial transformer layer to effectively alleviate the large view variation and learn location discriminative embeddings from the cross-view images. Furthermore, we propose to combine the triplet ranking loss with a simple and effective location identity loss to further enhance the performances. We test our method on a publicly available dataset and the results show that the proposed method outperforms state-of-the-art by a large margin.

KW - Cross-view geolocalization

KW - Deep metric learning

KW - Image retrieval

KW - Image-based localization

UR - http://www.scopus.com/inward/record.url?scp=85076884483&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-34120-6_5

DO - 10.1007/978-3-030-34120-6_5

M3 - Conference contribution

AN - SCOPUS:85076884483

SN - 9783030341190

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 57

EP - 67

BT - Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1

A2 - Zhao, Yao

A2 - Lin, Chunyu

A2 - Barnes, Nick

A2 - Chen, Baoquan

A2 - Westermann, Rüdiger

A2 - Kong, Xiangwei

PB - Springer

T2 - 10th International Conference on Image and Graphics, ICIG 2019

Y2 - 23 August 2019 through 25 August 2019

ER -

Cao R, Zhu J, Li Q, Zhang Q, Li Q, Liu B et al. Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization. In Zhao Y, Lin C, Barnes N, Chen B, Westermann R, Kong X, editors, Image and Graphics - 10th International Conference, ICIG 2019, Proceedings, Part 1. Springer. 2019. p. 57-67. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-34120-6_5

Learning Spatial-Aware Cross-View Embeddings for Ground-to-Aerial Geolocalization

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this