Detecting and discriminating behavioural anomalies

This paper aims to address the problem of anomaly detection and discrimination in complex behaviours, where anomalies are subtle and difficult to detect owing to the complex temporal dynamics and correlations among multiple objects' behaviours. Specifically, we decompose a complex behaviour pattern according to its temporal characteristics or spatial-temporal visual contexts. The decomposed behaviour is then modelled using a cascade of Dynamic Bayesian Networks (CasDBNs). In contrast to existing standalone models, the proposed behaviour decomposition and cascade modelling offers distinct advantage in simplicity for complex behaviour modelling. Importantly, the decomposition and cascade structure map naturally to the structure of complex behaviour, allowing for a more effective detection of subtle anomalies in surveillance videos. Comparative experiments using both indoor and outdoor data are carried out to demonstrate that, in addition to the novel capability of discriminating different types of anomalies, the proposed framework outperforms existing methods in detecting durational anomalies in complex behaviours and subtle anomalies that are difficult to detect when objects are viewed in isolation.