GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network

Peiran Peng; Feng Mu; Peilin Yan; Liqiang Song; Hui Li; Yu Chen; Jianan Li; Tingfa Xu

doi:10.1007/978-3-031-10464-0_35

GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network

Peiran Peng, Feng Mu, Peilin Yan, Liqiang Song, Hui Li, Yu Chen, Jianan Li, Tingfa Xu^*

^*Corresponding author for this work

School of Optics and Photonics

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

1 Citation (Scopus)

Abstract

Pedestrian detection is a critical but challenging research field widely applicable in self-driving, surveillance and robotics. The performance of pedestrian detection is not ideal under the limitation of imaging conditions, especially at night or occlusion. To overcome these obstacles, we propose a cross-modal pedestrian detection network based on Gaussian Cross Attention (GCANet) improving the detection performance by a full use of multi-modal features. Through the bidirectional coupling of local features of different modals, the feature interaction and fusion between different modals are realized, and the salient features between multi-modal are effectively emphasized, thus improving the detection accuracy. Experimental results demonstrate GCANet achieves the highest accuracy with the state-of-the-art on KAIST multi-modal pedestrian dataset.

Original language	English
Title of host publication	Intelligent Computing - Proceedings of the 2022 Computing Conference
Editors	Kohei Arai
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	520-530
Number of pages	11
ISBN (Print)	9783031104633
DOIs	https://doi.org/10.1007/978-3-031-10464-0_35
Publication status	Published - 2022
Event	Computing Conference, 2022 - Virtual, Online Duration: 14 Jul 2022 → 15 Jul 2022

Publication series

Name	Lecture Notes in Networks and Systems
Volume	507 LNNS
ISSN (Print)	2367-3370
ISSN (Electronic)	2367-3389

Conference

Conference	Computing Conference, 2022
City	Virtual, Online
Period	14/07/22 → 15/07/22

Keywords

Gaussian Cross Attention
Multi-modal fusion
Pedestrian detection

Access to Document

10.1007/978-3-031-10464-0_35

Cite this

Peng, P., Mu, F., Yan, P., Song, L., Li, H., Chen, Y., Li, J., & Xu, T. (2022). GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network. In K. Arai (Ed.), Intelligent Computing - Proceedings of the 2022 Computing Conference (pp. 520-530). (Lecture Notes in Networks and Systems; Vol. 507 LNNS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-10464-0_35

@inproceedings{248ce482caba4f049c0caccd13291d19,

title = "GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network",

abstract = "Pedestrian detection is a critical but challenging research field widely applicable in self-driving, surveillance and robotics. The performance of pedestrian detection is not ideal under the limitation of imaging conditions, especially at night or occlusion. To overcome these obstacles, we propose a cross-modal pedestrian detection network based on Gaussian Cross Attention (GCANet) improving the detection performance by a full use of multi-modal features. Through the bidirectional coupling of local features of different modals, the feature interaction and fusion between different modals are realized, and the salient features between multi-modal are effectively emphasized, thus improving the detection accuracy. Experimental results demonstrate GCANet achieves the highest accuracy with the state-of-the-art on KAIST multi-modal pedestrian dataset.",

keywords = "Gaussian Cross Attention, Multi-modal fusion, Pedestrian detection",

author = "Peiran Peng and Feng Mu and Peilin Yan and Liqiang Song and Hui Li and Yu Chen and Jianan Li and Tingfa Xu",

note = "Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; Computing Conference, 2022 ; Conference date: 14-07-2022 Through 15-07-2022",

year = "2022",

doi = "10.1007/978-3-031-10464-0\_35",

language = "English",

isbn = "9783031104633",

series = "Lecture Notes in Networks and Systems",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "520--530",

editor = "Kohei Arai",

booktitle = "Intelligent Computing - Proceedings of the 2022 Computing Conference",

address = "Germany",

}

Peng, P, Mu, F, Yan, P, Song, L, Li, H, Chen, Y, Li, J & Xu, T 2022, GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network. in K Arai (ed.), Intelligent Computing - Proceedings of the 2022 Computing Conference. Lecture Notes in Networks and Systems, vol. 507 LNNS, Springer Science and Business Media Deutschland GmbH, pp. 520-530, Computing Conference, 2022, Virtual, Online, 14/07/22. https://doi.org/10.1007/978-3-031-10464-0_35

GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network. / Peng, Peiran; Mu, Feng; Yan, Peilin et al.
Intelligent Computing - Proceedings of the 2022 Computing Conference. ed. / Kohei Arai. Springer Science and Business Media Deutschland GmbH, 2022. p. 520-530 (Lecture Notes in Networks and Systems; Vol. 507 LNNS).

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

TY - GEN

T1 - GCANet

T2 - Computing Conference, 2022

AU - Peng, Peiran

AU - Mu, Feng

AU - Yan, Peilin

AU - Song, Liqiang

AU - Li, Hui

AU - Chen, Yu

AU - Li, Jianan

AU - Xu, Tingfa

PY - 2022

Y1 - 2022

N2 - Pedestrian detection is a critical but challenging research field widely applicable in self-driving, surveillance and robotics. The performance of pedestrian detection is not ideal under the limitation of imaging conditions, especially at night or occlusion. To overcome these obstacles, we propose a cross-modal pedestrian detection network based on Gaussian Cross Attention (GCANet) improving the detection performance by a full use of multi-modal features. Through the bidirectional coupling of local features of different modals, the feature interaction and fusion between different modals are realized, and the salient features between multi-modal are effectively emphasized, thus improving the detection accuracy. Experimental results demonstrate GCANet achieves the highest accuracy with the state-of-the-art on KAIST multi-modal pedestrian dataset.

AB - Pedestrian detection is a critical but challenging research field widely applicable in self-driving, surveillance and robotics. The performance of pedestrian detection is not ideal under the limitation of imaging conditions, especially at night or occlusion. To overcome these obstacles, we propose a cross-modal pedestrian detection network based on Gaussian Cross Attention (GCANet) improving the detection performance by a full use of multi-modal features. Through the bidirectional coupling of local features of different modals, the feature interaction and fusion between different modals are realized, and the salient features between multi-modal are effectively emphasized, thus improving the detection accuracy. Experimental results demonstrate GCANet achieves the highest accuracy with the state-of-the-art on KAIST multi-modal pedestrian dataset.

KW - Gaussian Cross Attention

KW - Multi-modal fusion

KW - Pedestrian detection

UR - https://www.scopus.com/pages/publications/85135027095

U2 - 10.1007/978-3-031-10464-0_35

DO - 10.1007/978-3-031-10464-0_35

M3 - Conference contribution

AN - SCOPUS:85135027095

SN - 9783031104633

T3 - Lecture Notes in Networks and Systems

SP - 520

EP - 530

BT - Intelligent Computing - Proceedings of the 2022 Computing Conference

A2 - Arai, Kohei

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 14 July 2022 through 15 July 2022

ER -

Peng P, Mu F, Yan P, Song L, Li H, Chen Y et al. GCANet: A Cross-Modal Pedestrian Detection Method Based on Gaussian Cross Attention Network. In Arai K, editor, Intelligent Computing - Proceedings of the 2022 Computing Conference. Springer Science and Business Media Deutschland GmbH. 2022. p. 520-530. (Lecture Notes in Networks and Systems). doi: 10.1007/978-3-031-10464-0_35