EUVIMEDEUVIMED entdecken
Crossref

P‐42: Unsupervised Video Anomaly Detection with Vision–Language Model as Assistance

Yalong Jiang, Lian Huai, Yuyu Liu, Xingqun Jiang

SID Symposium Digest of Technical Papers · 2026-06 · 57 (1) · 1834-1837

DOI: 10.1002/sdtp.70001_498 ↗

Kurz gefasst

Fachartikel in der Zeitschrift SID Symposium Digest of Technical Papers, online erschienen am 11.09.2026. Verfasst von Yalong Jiang und 3 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Video anomaly detection (VAD) is crucial for security and smart‐city applications, but remains difficult due to the rarity, diversity, and weak definition of anomalous events, as well as expensive temporal annotation. Most existing methods rely on opaque visual embeddings and struggle to generalize across scenes and categories. We propose a logic‐guided vision–language framework that models anomalies as compositions of human‐interpretable cues. For each event type (accident, fighting, shooting, riot, explosion, abuse), we define direct, indirect, and counter cues and prompt a VLM to output calibrated cue scores and structured JSON decisions. A symbolic rule layer aggregates these scores into frame‐level anomaly predictions without training the backbone. This approach reduces label dependence, improves cross‐category transfer, and yields interpretable reasoning. On XD‐Violence dataset, our method achieves competitive performance.

Zitieren

APA

Jiang, Y., Huai, L., Liu, Y., & Jiang, X. (2026). P‐42: Unsupervised Video Anomaly Detection with Vision–Language Model as Assistance. SID Symposium Digest of Technical Papers, 57(1), 1834-1837. https://doi.org/10.1002/sdtp.70001_498

Vancouver

Jiang Y, Huai L, Liu Y, Jiang X. P‐42: Unsupervised Video Anomaly Detection with Vision–Language Model as Assistance. SID Symposium Digest of Technical Papers. 2026;57(1):1834-1837. doi:10.1002/sdtp.70001_498

BibTeX
@article{Jiang2026,
  author = {Jiang, Yalong and Huai, Lian and Liu, Yuyu and Jiang, Xingqun},
  title = {P‐42: Unsupervised Video Anomaly Detection with Vision–Language Model as Assistance},
  journal = {SID Symposium Digest of Technical Papers},
  year = {2026},
  volume = {57},
  number = {1},
  pages = {1834--1837},
  doi = {10.1002/sdtp.70001_498}
}
Kostenlos starten →