Citation

Yao Y, Wang X, Xu M, Pu Z, Wang Y, Atkins E, Crandall D. IEEE Trans. Pattern Anal. Mach. Intell. 2022; ePub(ePub): ePub.

Copyright

(Copyright © 2022, Institute of Electrical and Electronics Engineers, Publisher IEEE (Institute of Electrical and Electronics Engineers))

DOI

10.1109/TPAMI.2022.3150763

PMID

35157576

Abstract

Video anomaly detection has been extensively studied for static cameras but is much more challenging in egocentric driving videos where the scenes are extremely dynamic. This paper proposes an unsupervised method for anomaly detection based on future object localization. The idea is to predict locations of traffic participants short time steps into the future, and then monitor the accuracy and consistency of these predictions as evidence of an anomaly: inconsistent predictions tend to indicate that an anomaly has or is about to occur. To evaluate our method, we introduce a new large-scale benchmark dataset called Detection of Traffic Anomaly (DoTA) containing 4,677 videos with temporal, spatial, and categorical annotations. We also propose a new evaluation metric, which we call spatial-temporal area under curve (STAUC), and show that it captures how well a model detects both temporal and spatial locations of anomalous events (unlike existing metrics which focus only on temporal localization). Experimental results show that our method outperforms state-of-the-art methods on DoTA in terms of both metrics. In addition, we use the rich categorical annotations in DoTA to benchmark video action detection and online action detection methods. The DoTA dataset has been made available at: https://github.com/MoonBlvd/Detection-of-Traffic-Anomaly.

Language: en