Video Anomaly Detection

At KIST AI·Robotics, I study video anomaly detection in surveillance scenarios with a focus on modeling contextual dynamics — integrating RGB appearance, motion dynamics, and scene information so that the model better captures anomaly concepts through their relationships.

Recent work explores VLM-guided unsupervised approaches with adaptive sampling for efficiency and target-domain transfer.

References

2026

BMVC

BUSTER: Adaptive Sampling for VLM-Guided Unsupervised Video Anomaly Detection

Deokhyun Ahn, Yonghun Choi, Jae-Ho Han, Ig-Jae Kim, and Haksub Kim

In Proceedings of the British Machine Vision Conference (BMVC), 2026

Under Review
BMVC

Learning Robust Representations for Few-Shot Action Recognition with Frame-Level Ambiguities

Deokhyun Ahn, Yongjin Jo, Ui-Seok Lee, Haesol Park, Jae-Ho Han, and Haksub Kim

In Proceedings of the British Machine Vision Conference (BMVC), 2026

Under Review

2025

TIP

Where and What: Contextual Dynamics-Aware Anomaly Detection in Surveillance Videos

Deokhyun Ahn, Yongjin Jo, DongBum Kim, Gi Pyo Nam, Jae-Ho Han, and Haksub Kim

IEEE Transactions on Image Processing, 2025

IF 13.7 · top 2.2% in Electrical & Electronic Engineering

Bib HTML

@article{ahn2025where,
  title = {Where and What: Contextual Dynamics-Aware Anomaly Detection in Surveillance Videos},
  author = {Ahn, Deokhyun and Jo, Yongjin and Kim, DongBum and Nam, Gi Pyo and Han, Jae-Ho and Kim, Haksub},
  journal = {IEEE Transactions on Image Processing},
  year = {2025},
  note = {IF 13.7 · top 2.2\% in Electrical \& Electronic Engineering},
}

ESWA

A Foundational Research Framework for Real-World Abandoned Object Detection: Train-Free Baseline and a Standardized Benchmark

DongBum Kim, Deokhyun Ahn, Yongjin Jo, Haesol Park, Sangyoun Lee, and Haksub Kim

Expert Systems with Applications, 2025

Bib HTML

@article{kim2025abandoned,
  title = {A Foundational Research Framework for Real-World Abandoned Object Detection: Train-Free Baseline and a Standardized Benchmark},
  author = {Kim, DongBum and Ahn, Deokhyun and Jo, Yongjin and Park, Haesol and Lee, Sangyoun and Kim, Haksub},
  journal = {Expert Systems with Applications},
  pages = {130658},
  year = {2025}
}