MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos

Yiling Zhang; Erkut Akdag; Egor Bondarev; Peter H.N. de With

doi:10.1117/12.3055069

MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos

Yiling Zhang
, Erkut Akdag (Corresponding author)
, Egor Bondarev
, Peter H.N. de With

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

6 Citations (Scopus)

84 Downloads (Pure)

Abstract

Detection of anomaly events is relevant for public safety and requires a combination of fine-grained motion information and contextual events at variable time-scales. To this end, we propose a Multi-Timescale Feature Learning (MTFL) method to enhance the representation of anomaly features. Short, medium, and long temporal tubelets are employed to extract spatio-temporal video features using a Video Swin Transformer. Experimental results demonstrate that MTFL outperforms state-of-the-art methods on the UCF-Crime dataset, achieving an anomaly detection performance 89.78% AUC. Moreover, it performs complementary to SotA with 95.32% AUC on the ShanghaiTech and 84.57% AP on the XD-Violence dataset. Furthermore, we generate an extended dataset of the UCF-Crime for development and evaluation on a wider range of anomalies, namely Video Anomaly Detection Dataset (VADD), involving 2,591 videos in 18 classes with extensive coverage of realistic anomalies.

Original language	English
Title of host publication	Seventeenth International Conference on Machine Vision, ICMV 2024
Editors	Wolfgang Osten
Publisher	SPIE
Number of pages	8
ISBN (Electronic)	9781510688285
ISBN (Print)	9781510688278
DOIs	https://doi.org/10.1117/12.3055069
Publication status	Published - 24 Feb 2025
Event	17th International Conference on Machine Vision, ICMV 2024 - Edinburg, United Kingdom Duration: 10 Oct 2024 → 13 Oct 2024

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	13517
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	17th International Conference on Machine Vision, ICMV 2024
Country/Territory	United Kingdom
City	Edinburg
Period	10/10/24 → 13/10/24

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 16 Peace, Justice and Strong Institutions

Keywords

Anomaly detection
Surveillance videos
Video understanding

Access to Document

10.1117/12.3055069

pdfFinal published version, 743 KBLicence: TAVERNE

Cite this

Zhang, Y., Akdag, E., Bondarev, E., & de With, P. H. N. (2025). MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos. In W. Osten (Ed.), Seventeenth International Conference on Machine Vision, ICMV 2024 Article 135170M (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13517). SPIE. https://doi.org/10.1117/12.3055069

@inproceedings{224f1298327f494f8089d669aff4f061,

title = "MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos",

abstract = "Detection of anomaly events is relevant for public safety and requires a combination of fine-grained motion information and contextual events at variable time-scales. To this end, we propose a Multi-Timescale Feature Learning (MTFL) method to enhance the representation of anomaly features. Short, medium, and long temporal tubelets are employed to extract spatio-temporal video features using a Video Swin Transformer. Experimental results demonstrate that MTFL outperforms state-of-the-art methods on the UCF-Crime dataset, achieving an anomaly detection performance 89.78\% AUC. Moreover, it performs complementary to SotA with 95.32\% AUC on the ShanghaiTech and 84.57\% AP on the XD-Violence dataset. Furthermore, we generate an extended dataset of the UCF-Crime for development and evaluation on a wider range of anomalies, namely Video Anomaly Detection Dataset (VADD), involving 2,591 videos in 18 classes with extensive coverage of realistic anomalies.",

keywords = "Anomaly detection, Surveillance videos, Video understanding",

author = "Yiling Zhang and Erkut Akdag and Egor Bondarev and \{de With\}, \{Peter H.N.\}",

year = "2025",

month = feb,

day = "24",

doi = "10.1117/12.3055069",

language = "English",

isbn = "9781510688278",

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

publisher = "SPIE",

editor = "Wolfgang Osten",

booktitle = "Seventeenth International Conference on Machine Vision, ICMV 2024",

address = "United States",

note = "17th International Conference on Machine Vision, ICMV 2024 ; Conference date: 10-10-2024 Through 13-10-2024",

}

Zhang, Y, Akdag, E , Bondarev, E & de With, PHN 2025, MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos. in W Osten (ed.), Seventeenth International Conference on Machine Vision, ICMV 2024., 135170M, Proceedings of SPIE - The International Society for Optical Engineering, vol. 13517, SPIE, 17th International Conference on Machine Vision, ICMV 2024, Edinburg, United Kingdom, 10/10/24. https://doi.org/10.1117/12.3055069

MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos. / Zhang, Yiling; Akdag, Erkut (Corresponding author); Bondarev, Egor et al.
Seventeenth International Conference on Machine Vision, ICMV 2024. ed. / Wolfgang Osten. SPIE, 2025. 135170M (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13517).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - MTFL

T2 - 17th International Conference on Machine Vision, ICMV 2024

AU - Zhang, Yiling

AU - Akdag, Erkut

AU - Bondarev, Egor

AU - de With, Peter H.N.

PY - 2025/2/24

Y1 - 2025/2/24

N2 - Detection of anomaly events is relevant for public safety and requires a combination of fine-grained motion information and contextual events at variable time-scales. To this end, we propose a Multi-Timescale Feature Learning (MTFL) method to enhance the representation of anomaly features. Short, medium, and long temporal tubelets are employed to extract spatio-temporal video features using a Video Swin Transformer. Experimental results demonstrate that MTFL outperforms state-of-the-art methods on the UCF-Crime dataset, achieving an anomaly detection performance 89.78% AUC. Moreover, it performs complementary to SotA with 95.32% AUC on the ShanghaiTech and 84.57% AP on the XD-Violence dataset. Furthermore, we generate an extended dataset of the UCF-Crime for development and evaluation on a wider range of anomalies, namely Video Anomaly Detection Dataset (VADD), involving 2,591 videos in 18 classes with extensive coverage of realistic anomalies.

AB - Detection of anomaly events is relevant for public safety and requires a combination of fine-grained motion information and contextual events at variable time-scales. To this end, we propose a Multi-Timescale Feature Learning (MTFL) method to enhance the representation of anomaly features. Short, medium, and long temporal tubelets are employed to extract spatio-temporal video features using a Video Swin Transformer. Experimental results demonstrate that MTFL outperforms state-of-the-art methods on the UCF-Crime dataset, achieving an anomaly detection performance 89.78% AUC. Moreover, it performs complementary to SotA with 95.32% AUC on the ShanghaiTech and 84.57% AP on the XD-Violence dataset. Furthermore, we generate an extended dataset of the UCF-Crime for development and evaluation on a wider range of anomalies, namely Video Anomaly Detection Dataset (VADD), involving 2,591 videos in 18 classes with extensive coverage of realistic anomalies.

KW - Anomaly detection

KW - Surveillance videos

KW - Video understanding

UR - https://www.scopus.com/pages/publications/105000118808

UR - https://erktkdg.github.io/MTFL/

U2 - 10.1117/12.3055069

DO - 10.1117/12.3055069

M3 - Conference contribution

AN - SCOPUS:105000118808

SN - 9781510688278

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

BT - Seventeenth International Conference on Machine Vision, ICMV 2024

A2 - Osten, Wolfgang

PB - SPIE

Y2 - 10 October 2024 through 13 October 2024

ER -

MTFL: Multi-Timescale Feature Learning for Weakly-supervised Anomaly Detection in Surveillance Videos

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this