An Evolutionary Optimization based Interval Type-2 Fuzzy Classification System for Human Behaviour Recognition and Summarisation

Automatic recognition of behaviours and events from visual data is an emerging topic in video surveillance. These methods promise the ability to derive contextual awareness for a scene and may further enable the ability to predict the intentions of the subject. This paper describes a novel system for analysing human behaviours in the context of a video surveillance application. This may be used to distinguish between normal and anomalous behaviours. We propose a novel framework for the application of behaviour recognition and summarisation using interval type-2 fuzzy logic classification systems (IT2FLS). We employ the evolutionary-based technique Big Bang Big Crunch (BB-BC) to automatically optimise parameters of membership functions (MFs) and rules in the IT2FLSs. Our analysis shows that the BB-BC IT2FLS is able to robustly recognise behaviours and furthermore outperforms both its' conventional IT2FLS (which doesnot employ fuzzy classification techniques) and Type-1 FLSs (T1FLSs) counterparts in addition to non-fuzzy recognition methods.