Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection

Huamin Ren, Weifeng Liu, Søren Ingvor Olsen, Sergio Escalera, Thomas B. Moeslund

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

626 Downloads (Pure)

Resumé

Abnormal event detection has been an important issue in video surveillance applications. Due to the huge amount of surveillance data, only a small proportion could be loaded during the training. As a result, there is a high chance of incomplete normal
patterns in the training data, which makes the task very challenging. Sparse representation, as one of solutions, has shown its effectiveness. The basic principle is to find a collection (a dictionary) of atoms so that each training sample can only be represented
by a few atoms. However, the relationship of atoms within the dictionary is commonly neglected, which brings a high risk of false alarm rate: atoms from infrequent normal patterns are difficult to be distinguished from real anomalies. In this paper, we
propose behavior-specific dictionaries (BSD) through unsupervised learning, in which atoms from the same dictionary representing one type of normal behavior in the training video. Moreover, ‘missed atoms’ that are potentially from infrequent normal features are used to refine these behavior dictionaries. To further reduce false alarms, the detection of abnormal features is not only dependent on reconstruction error from the learned dictionaries, but also on non zero distribution in coefficients. Experimental results on Anomaly Stairs dataset and UCSD Anomaly dataset show the effectiveness of our algorithm. Remarkably, our BSD algorithm can improve AUC significantly by 10% on the stricter pixel-level evaluation, compared to the best result that has been reported so far.
OriginalsprogEngelsk
TitelProceedings of the British Machine Vision Conference (BMVC), 2015
RedaktørerXianghua Xie, Mark W. Jones, Gary K. L. Tam
Antal sider13
ForlagBritish Machine Vision Association
Publikationsdato2015
Sider28.1-28.13
ISBN (Trykt)1-901725-53-7
DOI
StatusUdgivet - 2015
BegivenhedBritish Machine Vision Conference 2015: Machine Vision of Animals and their Behaviour - Swansea University, Swansea, Storbritannien
Varighed: 7 sep. 201510 sep. 2015
Konferencens nummer: 26

Konference

KonferenceBritish Machine Vision Conference 2015
Nummer26
LokationSwansea University
LandStorbritannien
BySwansea
Periode07/09/201510/09/2015

Fingerprint

Glossaries
Atoms
Stairs
Unsupervised learning
Pixels

Citer dette

Ren, H., Liu, W., Olsen, S. I., Escalera, S., & Moeslund, T. B. (2015). Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection. I X. Xie, M. W. Jones, & G. K. L. Tam (red.), Proceedings of the British Machine Vision Conference (BMVC), 2015 (s. 28.1-28.13). British Machine Vision Association. https://doi.org/10.5244/C.29.28
Ren, Huamin ; Liu, Weifeng ; Olsen, Søren Ingvor ; Escalera, Sergio ; Moeslund, Thomas B. / Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection. Proceedings of the British Machine Vision Conference (BMVC), 2015. red. / Xianghua Xie ; Mark W. Jones ; Gary K. L. Tam. British Machine Vision Association, 2015. s. 28.1-28.13
@inproceedings{8b153a71eb2648de81b7e2fa639a40d0,
title = "Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection",
abstract = "Abnormal event detection has been an important issue in video surveillance applications. Due to the huge amount of surveillance data, only a small proportion could be loaded during the training. As a result, there is a high chance of incomplete normalpatterns in the training data, which makes the task very challenging. Sparse representation, as one of solutions, has shown its effectiveness. The basic principle is to find a collection (a dictionary) of atoms so that each training sample can only be representedby a few atoms. However, the relationship of atoms within the dictionary is commonly neglected, which brings a high risk of false alarm rate: atoms from infrequent normal patterns are difficult to be distinguished from real anomalies. In this paper, wepropose behavior-specific dictionaries (BSD) through unsupervised learning, in which atoms from the same dictionary representing one type of normal behavior in the training video. Moreover, ‘missed atoms’ that are potentially from infrequent normal features are used to refine these behavior dictionaries. To further reduce false alarms, the detection of abnormal features is not only dependent on reconstruction error from the learned dictionaries, but also on non zero distribution in coefficients. Experimental results on Anomaly Stairs dataset and UCSD Anomaly dataset show the effectiveness of our algorithm. Remarkably, our BSD algorithm can improve AUC significantly by 10{\%} on the stricter pixel-level evaluation, compared to the best result that has been reported so far.",
author = "Huamin Ren and Weifeng Liu and Olsen, {S{\o}ren Ingvor} and Sergio Escalera and Moeslund, {Thomas B.}",
year = "2015",
doi = "10.5244/C.29.28",
language = "English",
isbn = "1-901725-53-7",
pages = "28.1--28.13",
editor = "Xianghua Xie and Jones, {Mark W.} and Tam, {Gary K. L.}",
booktitle = "Proceedings of the British Machine Vision Conference (BMVC), 2015",
publisher = "British Machine Vision Association",

}

Ren, H, Liu, W, Olsen, SI, Escalera, S & Moeslund, TB 2015, Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection. i X Xie, MW Jones & GKL Tam (red), Proceedings of the British Machine Vision Conference (BMVC), 2015. British Machine Vision Association, s. 28.1-28.13, Swansea, Storbritannien, 07/09/2015. https://doi.org/10.5244/C.29.28

Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection. / Ren, Huamin; Liu, Weifeng; Olsen, Søren Ingvor; Escalera, Sergio; Moeslund, Thomas B.

Proceedings of the British Machine Vision Conference (BMVC), 2015. red. / Xianghua Xie; Mark W. Jones; Gary K. L. Tam. British Machine Vision Association, 2015. s. 28.1-28.13.

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

TY - GEN

T1 - Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection

AU - Ren, Huamin

AU - Liu, Weifeng

AU - Olsen, Søren Ingvor

AU - Escalera, Sergio

AU - Moeslund, Thomas B.

PY - 2015

Y1 - 2015

N2 - Abnormal event detection has been an important issue in video surveillance applications. Due to the huge amount of surveillance data, only a small proportion could be loaded during the training. As a result, there is a high chance of incomplete normalpatterns in the training data, which makes the task very challenging. Sparse representation, as one of solutions, has shown its effectiveness. The basic principle is to find a collection (a dictionary) of atoms so that each training sample can only be representedby a few atoms. However, the relationship of atoms within the dictionary is commonly neglected, which brings a high risk of false alarm rate: atoms from infrequent normal patterns are difficult to be distinguished from real anomalies. In this paper, wepropose behavior-specific dictionaries (BSD) through unsupervised learning, in which atoms from the same dictionary representing one type of normal behavior in the training video. Moreover, ‘missed atoms’ that are potentially from infrequent normal features are used to refine these behavior dictionaries. To further reduce false alarms, the detection of abnormal features is not only dependent on reconstruction error from the learned dictionaries, but also on non zero distribution in coefficients. Experimental results on Anomaly Stairs dataset and UCSD Anomaly dataset show the effectiveness of our algorithm. Remarkably, our BSD algorithm can improve AUC significantly by 10% on the stricter pixel-level evaluation, compared to the best result that has been reported so far.

AB - Abnormal event detection has been an important issue in video surveillance applications. Due to the huge amount of surveillance data, only a small proportion could be loaded during the training. As a result, there is a high chance of incomplete normalpatterns in the training data, which makes the task very challenging. Sparse representation, as one of solutions, has shown its effectiveness. The basic principle is to find a collection (a dictionary) of atoms so that each training sample can only be representedby a few atoms. However, the relationship of atoms within the dictionary is commonly neglected, which brings a high risk of false alarm rate: atoms from infrequent normal patterns are difficult to be distinguished from real anomalies. In this paper, wepropose behavior-specific dictionaries (BSD) through unsupervised learning, in which atoms from the same dictionary representing one type of normal behavior in the training video. Moreover, ‘missed atoms’ that are potentially from infrequent normal features are used to refine these behavior dictionaries. To further reduce false alarms, the detection of abnormal features is not only dependent on reconstruction error from the learned dictionaries, but also on non zero distribution in coefficients. Experimental results on Anomaly Stairs dataset and UCSD Anomaly dataset show the effectiveness of our algorithm. Remarkably, our BSD algorithm can improve AUC significantly by 10% on the stricter pixel-level evaluation, compared to the best result that has been reported so far.

UR - http://bmvc2015.swansea.ac.uk/proceedings/toc.html

U2 - 10.5244/C.29.28

DO - 10.5244/C.29.28

M3 - Article in proceeding

SN - 1-901725-53-7

SP - 28.1-28.13

BT - Proceedings of the British Machine Vision Conference (BMVC), 2015

A2 - Xie, Xianghua

A2 - Jones, Mark W.

A2 - Tam, Gary K. L.

PB - British Machine Vision Association

ER -

Ren H, Liu W, Olsen SI, Escalera S, Moeslund TB. Unsupervised Behavior-Specific Dictionary Learning for Abnormal Event Detection. I Xie X, Jones MW, Tam GKL, red., Proceedings of the British Machine Vision Conference (BMVC), 2015. British Machine Vision Association. 2015. s. 28.1-28.13 https://doi.org/10.5244/C.29.28