Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification

W. Ansar; M. M. Fraz; M. Shahzad; I. Gohar; S. Javed; S. K. Jung

doi:10.1007/978-3-030-13469-3_76

Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification

W. Ansar, M. M. Fraz, M. Shahzad, I. Gohar, S. Javed, S. K. Jung

Electrical Engineering

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

5 Scopus citations

Abstract

Person re-identification (re-ID), is the task of associating the relationship among the images of a person captured from different cameras with non-overlapping field of view. Fundamental and yet an open issue in re-ID is extraction of powerful features in low resolution surveillance videos. In order to solve this, a novel Two Stream Convolutional Recurrent model with Attentive pooling mechanism is presented for person re-ID in videos. Each stream of the model is a Siamese network which is aimed at extracting and matching most differentiated feature maps. Attentive pooling is used to select most informative video frames. The output of two streams is fused to formulate one combined feature map, which helps to deal with major challenges of re-ID e.g. pose and illumination variation, clutter background and occlusion. The proposed technique is evaluated on three challenging datasets: MARS, PRID-2011 and iLIDS-VID. Experimental evaluation shows that the proposed technique performs better than existing state-of-the-art supervised video based person re-ID models. The implementation is available at https://github.com/re-identification/Person_RE-ID.git.

Original language	British English
Title of host publication	Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings
Editors	Ruben Vera-Rodriguez, Julian Fierrez, Aythami Morales
Publisher	Springer Verlag
Pages	654-661
Number of pages	8
ISBN (Print)	9783030134686
DOIs	https://doi.org/10.1007/978-3-030-13469-3_76
State	Published - 2019
Event	23rd Iberoamerican Congress on Pattern Recognition, CIARP 2018 - Madrid, Spain Duration: 19 Nov 2018 → 22 Nov 2018

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11401 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	23rd Iberoamerican Congress on Pattern Recognition, CIARP 2018
Country/Territory	Spain
City	Madrid
Period	19/11/18 → 22/11/18

Keywords

Person re-identification
Spatial stream
Temporal stream

Access to Document

10.1007/978-3-030-13469-3_76

Cite this

Ansar, W., Fraz, M. M., Shahzad, M., Gohar, I., Javed, S., & Jung, S. K. (2019). Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification. In R. Vera-Rodriguez, J. Fierrez, & A. Morales (Eds.), Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings (pp. 654-661). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11401 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-13469-3_76

Ansar, W. ; Fraz, M. M. ; Shahzad, M. et al. / Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification. Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings. editor / Ruben Vera-Rodriguez ; Julian Fierrez ; Aythami Morales. Springer Verlag, 2019. pp. 654-661 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{fdfccff8cc4b4e6da0472be17ae14832,

title = "Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification",

abstract = "Person re-identification (re-ID), is the task of associating the relationship among the images of a person captured from different cameras with non-overlapping field of view. Fundamental and yet an open issue in re-ID is extraction of powerful features in low resolution surveillance videos. In order to solve this, a novel Two Stream Convolutional Recurrent model with Attentive pooling mechanism is presented for person re-ID in videos. Each stream of the model is a Siamese network which is aimed at extracting and matching most differentiated feature maps. Attentive pooling is used to select most informative video frames. The output of two streams is fused to formulate one combined feature map, which helps to deal with major challenges of re-ID e.g. pose and illumination variation, clutter background and occlusion. The proposed technique is evaluated on three challenging datasets: MARS, PRID-2011 and iLIDS-VID. Experimental evaluation shows that the proposed technique performs better than existing state-of-the-art supervised video based person re-ID models. The implementation is available at https://github.com/re-identification/Person_RE-ID.git.",

keywords = "Person re-identification, Spatial stream, Temporal stream",

author = "W. Ansar and Fraz, {M. M.} and M. Shahzad and I. Gohar and S. Javed and Jung, {S. K.}",

note = "Funding Information: Acknowledgements. This research was supported by development project of leading technology for future vehicle of the business of Daegu metropolitan city (No. 20180910). We are also thankful to NVIDIA Corporation for donating the TitanX GPU which is used in this research. Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2019.; 23rd Iberoamerican Congress on Pattern Recognition, CIARP 2018 ; Conference date: 19-11-2018 Through 22-11-2018",

year = "2019",

doi = "10.1007/978-3-030-13469-3_76",

language = "British English",

isbn = "9783030134686",

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

publisher = "Springer Verlag",

pages = "654--661",

editor = "Ruben Vera-Rodriguez and Julian Fierrez and Aythami Morales",

booktitle = "Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings",

address = "Germany",

}

Ansar, W, Fraz, MM, Shahzad, M, Gohar, I, Javed, S & Jung, SK 2019, Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification. in R Vera-Rodriguez, J Fierrez & A Morales (eds), Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11401 LNCS, Springer Verlag, pp. 654-661, 23rd Iberoamerican Congress on Pattern Recognition, CIARP 2018, Madrid, Spain, 19/11/18. https://doi.org/10.1007/978-3-030-13469-3_76

Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification. / Ansar, W.; Fraz, M. M.; Shahzad, M. et al.
Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings. ed. / Ruben Vera-Rodriguez; Julian Fierrez; Aythami Morales. Springer Verlag, 2019. p. 654-661 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11401 LNCS).

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

TY - GEN

T1 - Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification

AU - Ansar, W.

AU - Fraz, M. M.

AU - Shahzad, M.

AU - Gohar, I.

AU - Javed, S.

AU - Jung, S. K.

N1 - Funding Information: Acknowledgements. This research was supported by development project of leading technology for future vehicle of the business of Daegu metropolitan city (No. 20180910). We are also thankful to NVIDIA Corporation for donating the TitanX GPU which is used in this research. Publisher Copyright: © Springer Nature Switzerland AG 2019.

PY - 2019

Y1 - 2019

N2 - Person re-identification (re-ID), is the task of associating the relationship among the images of a person captured from different cameras with non-overlapping field of view. Fundamental and yet an open issue in re-ID is extraction of powerful features in low resolution surveillance videos. In order to solve this, a novel Two Stream Convolutional Recurrent model with Attentive pooling mechanism is presented for person re-ID in videos. Each stream of the model is a Siamese network which is aimed at extracting and matching most differentiated feature maps. Attentive pooling is used to select most informative video frames. The output of two streams is fused to formulate one combined feature map, which helps to deal with major challenges of re-ID e.g. pose and illumination variation, clutter background and occlusion. The proposed technique is evaluated on three challenging datasets: MARS, PRID-2011 and iLIDS-VID. Experimental evaluation shows that the proposed technique performs better than existing state-of-the-art supervised video based person re-ID models. The implementation is available at https://github.com/re-identification/Person_RE-ID.git.

AB - Person re-identification (re-ID), is the task of associating the relationship among the images of a person captured from different cameras with non-overlapping field of view. Fundamental and yet an open issue in re-ID is extraction of powerful features in low resolution surveillance videos. In order to solve this, a novel Two Stream Convolutional Recurrent model with Attentive pooling mechanism is presented for person re-ID in videos. Each stream of the model is a Siamese network which is aimed at extracting and matching most differentiated feature maps. Attentive pooling is used to select most informative video frames. The output of two streams is fused to formulate one combined feature map, which helps to deal with major challenges of re-ID e.g. pose and illumination variation, clutter background and occlusion. The proposed technique is evaluated on three challenging datasets: MARS, PRID-2011 and iLIDS-VID. Experimental evaluation shows that the proposed technique performs better than existing state-of-the-art supervised video based person re-ID models. The implementation is available at https://github.com/re-identification/Person_RE-ID.git.

KW - Person re-identification

KW - Spatial stream

KW - Temporal stream

UR - http://www.scopus.com/inward/record.url?scp=85063028349&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-13469-3_76

DO - 10.1007/978-3-030-13469-3_76

M3 - Conference contribution

AN - SCOPUS:85063028349

SN - 9783030134686

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

SP - 654

EP - 661

BT - Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings

A2 - Vera-Rodriguez, Ruben

A2 - Fierrez, Julian

A2 - Morales, Aythami

PB - Springer Verlag

T2 - 23rd Iberoamerican Congress on Pattern Recognition, CIARP 2018

Y2 - 19 November 2018 through 22 November 2018

ER -

Ansar W, Fraz MM, Shahzad M, Gohar I, Javed S, Jung SK. Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification. In Vera-Rodriguez R, Fierrez J, Morales A, editors, Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications - 23rd Iberoamerican Congress, CIARP 2018, Proceedings. Springer Verlag. 2019. p. 654-661. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-13469-3_76

Two stream deep CNN-RNN attentive pooling architecture for video-based person re-identification

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this