Experimental Evaluation of GAN-Based One-Class Anomaly Detection on Office Monitoring

Ning Dong, Yusuke Hatae, Muhammad Fikko Fadjrimiratno, Tetsu Matsukawa, Einoshin Suzuki

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Citations (Scopus)


In this paper, we test two anomaly detection methods based on Generative Adversarial Networks (GAN) on office monitoring including humans. GAN-based methods, especially those equipped with encoders and decoders, have shown impressive results in detecting new anomalies from images. We have been working on human monitoring in office environments with autonomous mobile robots and are motivated to incorporate the impressive, recent progress of GAN-based methods. Lawson et al.’s work tackled a similar problem of anomalous detection in an indoor, patrol trajectory environment with their patrolbot with a GAN-based method, though crucial differences such as the absence of humans exist for our purpose. We test a variant of their method, which we call FA-GAN here, as well as the cutting-edge method of GANomaly on our own robotic dataset. Motivated to employ such a method for a turnable Video Camera Recorder (VCR) placed at a fixed point, we also test the two methods for another dataset. Our experimental evaluation and subsequent analyses revealed interesting tendencies of the two methods including the effect of a missing normal image for GANomaly and their dependencies on the anomaly threshold.

Original languageEnglish
Title of host publicationFoundations of Intelligent Systems - 25th International Symposium, ISMIS 2020, Proceedings
EditorsDenis Helic, Martin Stettinger, Alexander Felfernig, Gerhard Leitner, Zbigniew W. Ras
PublisherSpringer Science and Business Media Deutschland GmbH
Number of pages11
ISBN (Print)9783030594909
Publication statusPublished - 2020
Event25th International Symposium on Methodologies for Intelligent Systems, ISMIS 2020 - Graz, Austria
Duration: Sept 23 2020Sept 25 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12117 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference25th International Symposium on Methodologies for Intelligent Systems, ISMIS 2020

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Experimental Evaluation of GAN-Based One-Class Anomaly Detection on Office Monitoring'. Together they form a unique fingerprint.

Cite this