Occlusion-aware pedestrian detection

Christos Apostolopoulos; Kamal Nasrollahi; M. Hsuan Yang; Mohammad N.S. Jahromi; Thomas B. Moeslund

doi:10.1117/12.2523107

Occlusion-aware pedestrian detection

Christos Apostolopoulos, Kamal Nasrollahi, M. Hsuan Yang, Mohammad N.S. Jahromi, Thomas B. Moeslund

Department of Computer Science

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

Abstract

Failure in pedestrian detection systems can be extremely crucial, specifically in driverless driving. In this paper, failures in pedestrian detectors are refined by re-evaluating the results of state of the art pedestrian detection systems, via a fully convolutional neural network. The network is trained on a number of datasets which include a custom designed occluded pedestrian dataset to address the problem of occlusion. Results show that when applying the proposed network, detectors can not only maintain their state of the art performance, but they even decrease average false positives rate per image, especially in the case where pedestrians are occluded.

Original language	English
Title of host publication	Eleventh International Conference on Machine Vision, ICMV 2018
Editors	Dmitry P. Nikolaev, Antanas Verikas, Petia Radeva, Jianhong Zhou
Publisher	SPIE
ISBN (Electronic)	9781510627482
DOIs	https://doi.org/10.1117/12.2523107
Publication status	Published - 2019
Event	11th International Conference on Machine Vision, ICMV 2018 - Munich, Germany Duration: 2018 Nov 1 → 2018 Nov 3

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	11041
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	11th International Conference on Machine Vision, ICMV 2018
Country	Germany
City	Munich
Period	18/11/1 → 18/11/3

All Science Journal Classification (ASJC) codes

Electronic, Optical and Magnetic Materials
Condensed Matter Physics
Computer Science Applications
Applied Mathematics
Electrical and Electronic Engineering

Access to Document

10.1117/12.2523107

Fingerprint Dive into the research topics of 'Occlusion-aware pedestrian detection'. Together they form a unique fingerprint.

Cite this

Apostolopoulos, C., Nasrollahi, K., Yang, M. H., Jahromi, M. N. S., & Moeslund, T. B. (2019). Occlusion-aware pedestrian detection. In D. P. Nikolaev, A. Verikas, P. Radeva, & J. Zhou (Eds.), Eleventh International Conference on Machine Vision, ICMV 2018 [110410I] (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 11041). SPIE. https://doi.org/10.1117/12.2523107

Apostolopoulos, Christos ; Nasrollahi, Kamal ; Yang, M. Hsuan ; Jahromi, Mohammad N.S. ; Moeslund, Thomas B. / Occlusion-aware pedestrian detection. Eleventh International Conference on Machine Vision, ICMV 2018. editor / Dmitry P. Nikolaev ; Antanas Verikas ; Petia Radeva ; Jianhong Zhou. SPIE, 2019. (Proceedings of SPIE - The International Society for Optical Engineering).

@inproceedings{e95673b9c7c447989de1d413e24a76bb,

title = "Occlusion-aware pedestrian detection",

abstract = "Failure in pedestrian detection systems can be extremely crucial, specifically in driverless driving. In this paper, failures in pedestrian detectors are refined by re-evaluating the results of state of the art pedestrian detection systems, via a fully convolutional neural network. The network is trained on a number of datasets which include a custom designed occluded pedestrian dataset to address the problem of occlusion. Results show that when applying the proposed network, detectors can not only maintain their state of the art performance, but they even decrease average false positives rate per image, especially in the case where pedestrians are occluded.",

author = "Christos Apostolopoulos and Kamal Nasrollahi and Yang, {M. Hsuan} and Jahromi, {Mohammad N.S.} and Moeslund, {Thomas B.}",

note = "Publisher Copyright: Copyright {\textcopyright} 2019 SPIE. Copyright: Copyright 2019 Elsevier B.V., All rights reserved.; 11th International Conference on Machine Vision, ICMV 2018 ; Conference date: 01-11-2018 Through 03-11-2018",

year = "2019",

doi = "10.1117/12.2523107",

language = "English",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Nikolaev, {Dmitry P.} and Antanas Verikas and Petia Radeva and Jianhong Zhou",

booktitle = "Eleventh International Conference on Machine Vision, ICMV 2018",

address = "United States",

}

Apostolopoulos, C, Nasrollahi, K, Yang, MH, Jahromi, MNS & Moeslund, TB 2019, Occlusion-aware pedestrian detection. in DP Nikolaev, A Verikas, P Radeva & J Zhou (eds), Eleventh International Conference on Machine Vision, ICMV 2018., 110410I, Proceedings of SPIE - The International Society for Optical Engineering, vol. 11041, SPIE, 11th International Conference on Machine Vision, ICMV 2018, Munich, Germany, 18/11/1. https://doi.org/10.1117/12.2523107

Occlusion-aware pedestrian detection. / Apostolopoulos, Christos; Nasrollahi, Kamal; Yang, M. Hsuan; Jahromi, Mohammad N.S.; Moeslund, Thomas B.

Eleventh International Conference on Machine Vision, ICMV 2018. ed. / Dmitry P. Nikolaev; Antanas Verikas; Petia Radeva; Jianhong Zhou. SPIE, 2019. 110410I (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 11041).

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

TY - GEN

T1 - Occlusion-aware pedestrian detection

AU - Apostolopoulos, Christos

AU - Nasrollahi, Kamal

AU - Yang, M. Hsuan

AU - Jahromi, Mohammad N.S.

AU - Moeslund, Thomas B.

PY - 2019

Y1 - 2019

N2 - Failure in pedestrian detection systems can be extremely crucial, specifically in driverless driving. In this paper, failures in pedestrian detectors are refined by re-evaluating the results of state of the art pedestrian detection systems, via a fully convolutional neural network. The network is trained on a number of datasets which include a custom designed occluded pedestrian dataset to address the problem of occlusion. Results show that when applying the proposed network, detectors can not only maintain their state of the art performance, but they even decrease average false positives rate per image, especially in the case where pedestrians are occluded.

AB - Failure in pedestrian detection systems can be extremely crucial, specifically in driverless driving. In this paper, failures in pedestrian detectors are refined by re-evaluating the results of state of the art pedestrian detection systems, via a fully convolutional neural network. The network is trained on a number of datasets which include a custom designed occluded pedestrian dataset to address the problem of occlusion. Results show that when applying the proposed network, detectors can not only maintain their state of the art performance, but they even decrease average false positives rate per image, especially in the case where pedestrians are occluded.

UR - http://www.scopus.com/inward/record.url?scp=85063424847&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85063424847&partnerID=8YFLogxK

U2 - 10.1117/12.2523107

DO - 10.1117/12.2523107

M3 - Conference contribution

AN - SCOPUS:85063424847

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - Eleventh International Conference on Machine Vision, ICMV 2018

A2 - Nikolaev, Dmitry P.

A2 - Verikas, Antanas

A2 - Radeva, Petia

A2 - Zhou, Jianhong

PB - SPIE

T2 - 11th International Conference on Machine Vision, ICMV 2018

Y2 - 1 November 2018 through 3 November 2018

ER -