Hybrid Video Anomaly Detection for Anomalous Scenarios in Autonomous Driving

TL;DR

This paper introduces HF$^2$-VAD$_{AD}$, a novel video anomaly detection method tailored for autonomous driving, focusing on temporal context and pixel-wise anomaly detection in critical scenarios.

Contribution

It adapts a surveillance anomaly detection approach to autonomous driving, learning normality from an ego perspective for better detection of rare events.

Findings

Effective pixel-wise anomaly detection in autonomous driving scenarios

Learned normality representations improve detection accuracy

Applicable to rare and critical driving situations

Abstract

In autonomous driving, the most challenging scenarios can only be detected within their temporal context. Most video anomaly detection approaches focus either on surveillance or traffic accidents, which are only a subfield of autonomous driving. We present HF$^2$-VAD$_{AD}$, a variation of the HF$^2$-VAD surveillance video anomaly detection method for autonomous driving. We learn a representation of normality from a vehicle's ego perspective and evaluate pixel-wise anomaly detections in rare and critical scenarios.