A group of customers can be modeled an a collection of autonomous agents navigating according to a set of simple rules. We apply two consequent stages of clustering: first to obtain short term swarming events and later to group the events into higher level swarming activities.

Swarming events in space-time (axis t represents frame number of event; X, Y - average event location on the floor plane) Dot size corresponds to the validity of event. Dots of matching color belong to the same activity