Anomaly Detection

Cross-sensor monitoring that fuses CCTV, depth, thermal, audio, and biosignal feeds to flag safety-critical events before they escalate.

Self-/unsupervised embeddings that learn "normal" per ward, lab, or wearable fleet
Edge-friendly deployment on Jetson, ARM MCUs, and neuromorphic co-processors
Clinician-facing dashboards with traceable alerts and feedback loops

Research Area Overview

Anomaly Detection links together our historical efforts in data mining, healthcare analytics, wearable sensing, and IoT. Whether the data source is a microvascular image, a sleep-stage hypnogram, or an industrial sensor stream, our pipelines learn what “normal” looks like and quickly raise alerts when deviations appear.

Focus Topics

Multimodal anomaly and novelty detection for smart hospitals and edge devices
Signal-quality estimation for ECG/PPG wearables
Digital twin simulations that generate rare-but-critical failure cases

Multimodal Anomaly Detection Platform

Self-supervised representations combine CCTV, depth, audio, and biomedical sensors to detect rare events while preserving privacy.

Dongwook Kwon, Jaeyong Kim, Hyeonwoo Kim, Cheolsoo Park