212 Downloads (Pure)

Abstract

Person re-identification is a topic which has potential to be used for applications within forensics, flow analysis and queue monitoring. It is the process of matching persons across two or more camera views, most often by extracting colour and texture based hand-crafted features, to identify similar persons. Because of challenges regarding changes in lighting between views, occlusion
or even privacy issues, more focus has turned to overhead and depth based camera solutions. Therefore, we have developed a system, based on a Convolutional Neural Network (CNN) which is trained using both depth and RGB modalities to provide a fused feature. By training on a locally collected dataset, we achieve a rank-1 accuracy of 74.69%, increased by 16.00% compared to using a single modality. Furthermore, tests on two similar publicly available benchmark datasets of TVPR and DPI-T show accuracies of 77.66% and 90.36%, respectively, outperforming state-of-the-art results by 3.60% and 5.20%, respectively.
Original languageEnglish
Title of host publication16th International Conference of the Biometrics Special Interest Group
PublisherIEEE
Publication date2017
ISBN (Print)978-1-5386-0396-3
ISBN (Electronic)978-3-88579-664-0
DOIs
Publication statusPublished - 2017
Event16th International Conference of the Biometrics Special Interest Group - Darmstadt, Germany
Duration: 20 Sep 201722 Sep 2017

Conference

Conference16th International Conference of the Biometrics Special Interest Group
CountryGermany
CityDarmstadt
Period20/09/201722/09/2017
SeriesLecture Notes in Informatics
Volume2017
ISSN1617-5468

Fingerprint

Cameras
Neural networks
Textures
Lighting
Color
Monitoring

Keywords

  • Multimodal
  • Person Re-identification
  • Convolutional Neural Networks
  • Feature Fusion

Cite this

Lejbølle, A. R., Nasrollahi, K., Krogh, B., & Moeslund, T. B. (2017). Multimodal Neural Network for Overhead Person Re-identification. In 16th International Conference of the Biometrics Special Interest Group IEEE. Lecture Notes in Informatics, Vol.. 2017 https://doi.org/10.23919/BIOSIG.2017.8053514
Lejbølle, Aske Rasch ; Nasrollahi, Kamal ; Krogh, Benjamin ; Moeslund, Thomas B. / Multimodal Neural Network for Overhead Person Re-identification. 16th International Conference of the Biometrics Special Interest Group. IEEE, 2017. (Lecture Notes in Informatics, Vol. 2017).
@inproceedings{76f7bd0bcf544cbda1bae586d6dd4b12,
title = "Multimodal Neural Network for Overhead Person Re-identification",
abstract = "Person re-identification is a topic which has potential to be used for applications within forensics, flow analysis and queue monitoring. It is the process of matching persons across two or more camera views, most often by extracting colour and texture based hand-crafted features, to identify similar persons. Because of challenges regarding changes in lighting between views, occlusionor even privacy issues, more focus has turned to overhead and depth based camera solutions. Therefore, we have developed a system, based on a Convolutional Neural Network (CNN) which is trained using both depth and RGB modalities to provide a fused feature. By training on a locally collected dataset, we achieve a rank-1 accuracy of 74.69{\%}, increased by 16.00{\%} compared to using a single modality. Furthermore, tests on two similar publicly available benchmark datasets of TVPR and DPI-T show accuracies of 77.66{\%} and 90.36{\%}, respectively, outperforming state-of-the-art results by 3.60{\%} and 5.20{\%}, respectively.",
keywords = "Multimodal, Person Re-identification, Convolutional Neural Networks, Feature Fusion",
author = "Lejb{\o}lle, {Aske Rasch} and Kamal Nasrollahi and Benjamin Krogh and Moeslund, {Thomas B.}",
year = "2017",
doi = "10.23919/BIOSIG.2017.8053514",
language = "English",
isbn = "978-1-5386-0396-3",
series = "Lecture Notes in Informatics",
publisher = "IEEE",
booktitle = "16th International Conference of the Biometrics Special Interest Group",
address = "United States",

}

Lejbølle, AR, Nasrollahi, K, Krogh, B & Moeslund, TB 2017, Multimodal Neural Network for Overhead Person Re-identification. in 16th International Conference of the Biometrics Special Interest Group. IEEE, Lecture Notes in Informatics, vol. 2017, Darmstadt, Germany, 20/09/2017. https://doi.org/10.23919/BIOSIG.2017.8053514

Multimodal Neural Network for Overhead Person Re-identification. / Lejbølle, Aske Rasch; Nasrollahi, Kamal; Krogh, Benjamin; Moeslund, Thomas B.

16th International Conference of the Biometrics Special Interest Group. IEEE, 2017. (Lecture Notes in Informatics, Vol. 2017).

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

TY - GEN

T1 - Multimodal Neural Network for Overhead Person Re-identification

AU - Lejbølle, Aske Rasch

AU - Nasrollahi, Kamal

AU - Krogh, Benjamin

AU - Moeslund, Thomas B.

PY - 2017

Y1 - 2017

N2 - Person re-identification is a topic which has potential to be used for applications within forensics, flow analysis and queue monitoring. It is the process of matching persons across two or more camera views, most often by extracting colour and texture based hand-crafted features, to identify similar persons. Because of challenges regarding changes in lighting between views, occlusionor even privacy issues, more focus has turned to overhead and depth based camera solutions. Therefore, we have developed a system, based on a Convolutional Neural Network (CNN) which is trained using both depth and RGB modalities to provide a fused feature. By training on a locally collected dataset, we achieve a rank-1 accuracy of 74.69%, increased by 16.00% compared to using a single modality. Furthermore, tests on two similar publicly available benchmark datasets of TVPR and DPI-T show accuracies of 77.66% and 90.36%, respectively, outperforming state-of-the-art results by 3.60% and 5.20%, respectively.

AB - Person re-identification is a topic which has potential to be used for applications within forensics, flow analysis and queue monitoring. It is the process of matching persons across two or more camera views, most often by extracting colour and texture based hand-crafted features, to identify similar persons. Because of challenges regarding changes in lighting between views, occlusionor even privacy issues, more focus has turned to overhead and depth based camera solutions. Therefore, we have developed a system, based on a Convolutional Neural Network (CNN) which is trained using both depth and RGB modalities to provide a fused feature. By training on a locally collected dataset, we achieve a rank-1 accuracy of 74.69%, increased by 16.00% compared to using a single modality. Furthermore, tests on two similar publicly available benchmark datasets of TVPR and DPI-T show accuracies of 77.66% and 90.36%, respectively, outperforming state-of-the-art results by 3.60% and 5.20%, respectively.

KW - Multimodal

KW - Person Re-identification

KW - Convolutional Neural Networks

KW - Feature Fusion

U2 - 10.23919/BIOSIG.2017.8053514

DO - 10.23919/BIOSIG.2017.8053514

M3 - Article in proceeding

SN - 978-1-5386-0396-3

T3 - Lecture Notes in Informatics

BT - 16th International Conference of the Biometrics Special Interest Group

PB - IEEE

ER -

Lejbølle AR, Nasrollahi K, Krogh B, Moeslund TB. Multimodal Neural Network for Overhead Person Re-identification. In 16th International Conference of the Biometrics Special Interest Group. IEEE. 2017. (Lecture Notes in Informatics, Vol. 2017). https://doi.org/10.23919/BIOSIG.2017.8053514