We present a system for recognising human behaviour given a symbolic representation of surveillance videos. The input of our system is a set of time-stamped short-term behaviours, that is, behaviours taking place in a short period of time -- walking, running, standing still, etc -- detected on video frames. The output of our system is a set of recognised long-term behaviours -- fighting, leaving an object, collapsing, etc -- which are pre-defined temporal combinations of short-term behaviours. We develop a logic programming implementation of the Event Calculus to express the constraints on the short-term behaviours that, if satisfied, lead to the recognition of a long-term behaviour. We present experimental results concerning videos with several humans and objects, temporally overlapping and repetitive behaviours. Moreover, we show that our approach, which is not restricted to video surveillance, is better suited to a class of recognition applications than state-of-the-art recognition systems.
View on arXiv