Semantic Segmentation of Fisheye Images

Publikation: Beitrag in Buch/Bericht/Sammelwerk/KonferenzbandAufsatz in KonferenzbandForschungPeer-Review

Autoren

  • Gregor Blott
  • Masato Takami
  • Christian Heipke

Externe Organisationen

  • Robert Bosch GmbH
Forschungs-netzwerk anzeigen

Details

OriginalspracheEnglisch
Titel des SammelwerksComputer Vision
UntertitelECCV 2018 Workshops, Proceedings
Herausgeber/-innenLaura Leal-Taixé, Stefan Roth
ErscheinungsortCham
Herausgeber (Verlag)Springer Verlag
Seiten181-196
Seitenumfang16
Auflage1.
ISBN (elektronisch)9783030110093
ISBN (Print)9783030110086
PublikationsstatusVeröffentlicht - 23 Jan. 2019
Veranstaltung15th European Conference on Computer Vision, ECCV 2018 - Munich, Deutschland
Dauer: 8 Sept. 201814 Sept. 2018

Publikationsreihe

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Band11129 LNCS
ISSN (Print)0302-9743
ISSN (elektronisch)1611-3349

Abstract

Semantic segmentation of fisheye images (e.g., from action-cameras or smartphones) requires different training approaches and data than those of rectilinear images obtained using central projection. The shape of objects is distorted depending on the distance between the principal point and the object position in the image. Therefore, classical semantic segmentation approaches fall short in terms of performance compared to rectilinear data. A potential solution to this problem is the recording and annotation of a new dataset, however this is expensive and tedious. In this study, an alternative approach that modifies the augmentation stage of deep learning training to re-use rectilinear training data is presented. In this way we obtain a considerably higher semantic segmentation performance on the fisheye images: +18.3% intersection over union (IoU) for action-camera test images, +8.3% IoU for artificially generated fisheye data, and +18.0% IoU for challenging security scenes acquired in bird’s eye view.

ASJC Scopus Sachgebiete

Zitieren

Semantic Segmentation of Fisheye Images. / Blott, Gregor; Takami, Masato; Heipke, Christian.
Computer Vision: ECCV 2018 Workshops, Proceedings. Hrsg. / Laura Leal-Taixé; Stefan Roth. 1. Aufl. Cham: Springer Verlag, 2019. S. 181-196 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 11129 LNCS).

Publikation: Beitrag in Buch/Bericht/Sammelwerk/KonferenzbandAufsatz in KonferenzbandForschungPeer-Review

Blott, G, Takami, M & Heipke, C 2019, Semantic Segmentation of Fisheye Images. in L Leal-Taixé & S Roth (Hrsg.), Computer Vision: ECCV 2018 Workshops, Proceedings. 1. Aufl., Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bd. 11129 LNCS, Springer Verlag, Cham, S. 181-196, 15th European Conference on Computer Vision, ECCV 2018, Munich, Deutschland, 8 Sept. 2018. https://doi.org/10.1007/978-3-030-11009-3_10
Blott, G., Takami, M., & Heipke, C. (2019). Semantic Segmentation of Fisheye Images. In L. Leal-Taixé, & S. Roth (Hrsg.), Computer Vision: ECCV 2018 Workshops, Proceedings (1. Aufl., S. 181-196). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 11129 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-11009-3_10
Blott G, Takami M, Heipke C. Semantic Segmentation of Fisheye Images. in Leal-Taixé L, Roth S, Hrsg., Computer Vision: ECCV 2018 Workshops, Proceedings. 1. Aufl. Cham: Springer Verlag. 2019. S. 181-196. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-11009-3_10
Blott, Gregor ; Takami, Masato ; Heipke, Christian. / Semantic Segmentation of Fisheye Images. Computer Vision: ECCV 2018 Workshops, Proceedings. Hrsg. / Laura Leal-Taixé ; Stefan Roth. 1. Aufl. Cham : Springer Verlag, 2019. S. 181-196 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
Download
@inproceedings{b6900b3cdde64a699f78561e53bdf0ae,
title = "Semantic Segmentation of Fisheye Images",
abstract = "Semantic segmentation of fisheye images (e.g., from action-cameras or smartphones) requires different training approaches and data than those of rectilinear images obtained using central projection. The shape of objects is distorted depending on the distance between the principal point and the object position in the image. Therefore, classical semantic segmentation approaches fall short in terms of performance compared to rectilinear data. A potential solution to this problem is the recording and annotation of a new dataset, however this is expensive and tedious. In this study, an alternative approach that modifies the augmentation stage of deep learning training to re-use rectilinear training data is presented. In this way we obtain a considerably higher semantic segmentation performance on the fisheye images: +18.3% intersection over union (IoU) for action-camera test images, +8.3% IoU for artificially generated fisheye data, and +18.0% IoU for challenging security scenes acquired in bird{\textquoteright}s eye view.",
keywords = "Deep learning, Fisheye images, Semantic segmentation",
author = "Gregor Blott and Masato Takami and Christian Heipke",
year = "2019",
month = jan,
day = "23",
doi = "10.1007/978-3-030-11009-3_10",
language = "English",
isbn = "9783030110086",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "181--196",
editor = "Laura Leal-Taix{\'e} and Stefan Roth",
booktitle = "Computer Vision",
address = "Germany",
edition = "1.",
note = "15th European Conference on Computer Vision, ECCV 2018 ; Conference date: 08-09-2018 Through 14-09-2018",

}

Download

TY - GEN

T1 - Semantic Segmentation of Fisheye Images

AU - Blott, Gregor

AU - Takami, Masato

AU - Heipke, Christian

PY - 2019/1/23

Y1 - 2019/1/23

N2 - Semantic segmentation of fisheye images (e.g., from action-cameras or smartphones) requires different training approaches and data than those of rectilinear images obtained using central projection. The shape of objects is distorted depending on the distance between the principal point and the object position in the image. Therefore, classical semantic segmentation approaches fall short in terms of performance compared to rectilinear data. A potential solution to this problem is the recording and annotation of a new dataset, however this is expensive and tedious. In this study, an alternative approach that modifies the augmentation stage of deep learning training to re-use rectilinear training data is presented. In this way we obtain a considerably higher semantic segmentation performance on the fisheye images: +18.3% intersection over union (IoU) for action-camera test images, +8.3% IoU for artificially generated fisheye data, and +18.0% IoU for challenging security scenes acquired in bird’s eye view.

AB - Semantic segmentation of fisheye images (e.g., from action-cameras or smartphones) requires different training approaches and data than those of rectilinear images obtained using central projection. The shape of objects is distorted depending on the distance between the principal point and the object position in the image. Therefore, classical semantic segmentation approaches fall short in terms of performance compared to rectilinear data. A potential solution to this problem is the recording and annotation of a new dataset, however this is expensive and tedious. In this study, an alternative approach that modifies the augmentation stage of deep learning training to re-use rectilinear training data is presented. In this way we obtain a considerably higher semantic segmentation performance on the fisheye images: +18.3% intersection over union (IoU) for action-camera test images, +8.3% IoU for artificially generated fisheye data, and +18.0% IoU for challenging security scenes acquired in bird’s eye view.

KW - Deep learning

KW - Fisheye images

KW - Semantic segmentation

UR - http://www.scopus.com/inward/record.url?scp=85061724543&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-11009-3_10

DO - 10.1007/978-3-030-11009-3_10

M3 - Conference contribution

AN - SCOPUS:85061724543

SN - 9783030110086

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

SP - 181

EP - 196

BT - Computer Vision

A2 - Leal-Taixé, Laura

A2 - Roth, Stefan

PB - Springer Verlag

CY - Cham

T2 - 15th European Conference on Computer Vision, ECCV 2018

Y2 - 8 September 2018 through 14 September 2018

ER -