Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions

Publikation: Beitrag in FachzeitschriftArtikelForschungPeer-Review

Autoren

  • Gregor Blott
  • Jie Yu
  • Christian Heipke

Externe Organisationen

  • Robert Bosch GmbH
Forschungs-netzwerk anzeigen

Details

Titel in ÜbersetzungPersonenwiedererkennung in einem Fischaugenkameranetzwerk mit unterschiedlichen Blickrichtungen
OriginalspracheEnglisch
Seiten (von - bis)263-274
Seitenumfang12
FachzeitschriftPFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science
Jahrgang87
Ausgabenummer5-6
PublikationsstatusVeröffentlicht - 28 Okt. 2019

Abstract

Personenwiedererkennung ist eine entscheidende Komponente für unterschiedliche Anwendungen von Kameranetzen wie beispielsweise der
Sicherheitskamera- und der Automatisierungstechnik sowie der Analyse von Kaufverhalten in Geschäften. Trotz großer Fortschritte in den letzten Jahren ist die Wiedererkennungsleistung für praktische Anwendungen aufgrund der hohen Variabilität des Aussehens von Personen und einer hoher Szenenkomplexität inklusive Unterschieden in Blickrichtung, Skalierung und Beleuchtung noch nicht ausreichend gelöst. Wir stellen in diesem Beitrag ein neues Verfahren für die Personenwiedererkennung vor, das Bilder von Fischaugenkameras verwendet, die an der Decke montiert sind und in Nadirichtung schauen. Dabei werden Personen aus unterschiedlichen Ansichten aufgenommen. Um die verschiedenen Ansichten zu prozessieren, wird auf Basis von geometrischer Sensormodellierung und Deep Learning eine generische Verarbeitungskette genutzt. Das Verfahren wird auf einem angepassten öffentlich zugänglichen Datensatz, Market-1501, und einem neu erstellten Fischaugendatensatz evaluiert. Signifikante Verbesserungen werden in unseren Experimenten gezeigt. Der Ansatz erreicht mehr als 97% Rang#1 Wiedererkennungsleistung und eine Verbesserung von ca. 12% auf dem neu erstellten Fischaugendatensatz im
Vergleich zur Fusion mit zufälligen Blickrichtungen.

ASJC Scopus Sachgebiete

Zitieren

Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions. / Blott, Gregor; Yu, Jie; Heipke, Christian.
in: PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science, Jahrgang 87, Nr. 5-6, 28.10.2019, S. 263-274.

Publikation: Beitrag in FachzeitschriftArtikelForschungPeer-Review

Blott, G, Yu, J & Heipke, C 2019, 'Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions', PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science, Jg. 87, Nr. 5-6, S. 263-274. https://doi.org/10.1007/s41064-019-00083-y
Blott, G., Yu, J., & Heipke, C. (2019). Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions. PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science, 87(5-6), 263-274. https://doi.org/10.1007/s41064-019-00083-y
Blott G, Yu J, Heipke C. Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions. PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science. 2019 Okt 28;87(5-6):263-274. doi: 10.1007/s41064-019-00083-y
Blott, Gregor ; Yu, Jie ; Heipke, Christian. / Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions. in: PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science. 2019 ; Jahrgang 87, Nr. 5-6. S. 263-274.
Download
@article{3e96f24e1f314574b4757e21f5a13cd5,
title = "Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions",
abstract = "Person re-identification is a crucial component for multi-camera networks in different real-world applications such as surveillance, automation, and business analytics. Despite considerable recent progress, the performance in practice is still not satisfactory due to the high intra-person variation and significant complexity of the task, including differences in scale, viewing direction, and illumination. We propose a novel approach for person re-identification, which exploits multi-view information of a fisheye camera looking downwards from the ceiling. To handle this highly variable multi-view information, we build a generic pipeline for processing fisheye camera imagery based on geometric sensor modelling and deep learning. The proposed approach is evaluated on a re-mapped version of the publicly available Market-1501 dataset, and, in addition, on a new fisheye dataset. Significant improvements are shown in our experiments: our approach achieves more than 97% rank#1 recognition rate if applied on the re-mapped Market-1501 dataset; on the new fisheye dataset we find an improvement of about 12% compared to random-view fusion.",
keywords = "Deep learning, Fisheye, Multi-view, Person re-identification",
author = "Gregor Blott and Jie Yu and Christian Heipke",
year = "2019",
month = oct,
day = "28",
doi = "10.1007/s41064-019-00083-y",
language = "English",
volume = "87",
pages = "263--274",
number = "5-6",

}

Download

TY - JOUR

T1 - Multi-view Person Re-identification in a Fisheye Camera Network with Different Viewing Directions

AU - Blott, Gregor

AU - Yu, Jie

AU - Heipke, Christian

PY - 2019/10/28

Y1 - 2019/10/28

N2 - Person re-identification is a crucial component for multi-camera networks in different real-world applications such as surveillance, automation, and business analytics. Despite considerable recent progress, the performance in practice is still not satisfactory due to the high intra-person variation and significant complexity of the task, including differences in scale, viewing direction, and illumination. We propose a novel approach for person re-identification, which exploits multi-view information of a fisheye camera looking downwards from the ceiling. To handle this highly variable multi-view information, we build a generic pipeline for processing fisheye camera imagery based on geometric sensor modelling and deep learning. The proposed approach is evaluated on a re-mapped version of the publicly available Market-1501 dataset, and, in addition, on a new fisheye dataset. Significant improvements are shown in our experiments: our approach achieves more than 97% rank#1 recognition rate if applied on the re-mapped Market-1501 dataset; on the new fisheye dataset we find an improvement of about 12% compared to random-view fusion.

AB - Person re-identification is a crucial component for multi-camera networks in different real-world applications such as surveillance, automation, and business analytics. Despite considerable recent progress, the performance in practice is still not satisfactory due to the high intra-person variation and significant complexity of the task, including differences in scale, viewing direction, and illumination. We propose a novel approach for person re-identification, which exploits multi-view information of a fisheye camera looking downwards from the ceiling. To handle this highly variable multi-view information, we build a generic pipeline for processing fisheye camera imagery based on geometric sensor modelling and deep learning. The proposed approach is evaluated on a re-mapped version of the publicly available Market-1501 dataset, and, in addition, on a new fisheye dataset. Significant improvements are shown in our experiments: our approach achieves more than 97% rank#1 recognition rate if applied on the re-mapped Market-1501 dataset; on the new fisheye dataset we find an improvement of about 12% compared to random-view fusion.

KW - Deep learning

KW - Fisheye

KW - Multi-view

KW - Person re-identification

UR - http://www.scopus.com/inward/record.url?scp=85074700731&partnerID=8YFLogxK

U2 - 10.1007/s41064-019-00083-y

DO - 10.1007/s41064-019-00083-y

M3 - Article

AN - SCOPUS:85074700731

VL - 87

SP - 263

EP - 274

JO - PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science

JF - PFG - Journal of Photogrammetry, Remote Sensing and Geoinformation Science

SN - 2512-2789

IS - 5-6

ER -