This paper describes a real time system for tracking people using colour cues. People are tracked as they enter, move about and interact with one another in the monitored scene. Those who have re-entered the scene are also identified. The implemented tracker uses a simple yet effective technique allowing unsupervised background modelling based on video sequences that may contain foreground objects (people). A simple extension of the technique allows the automatic reconstruction of the background model when drastic changes occur due to accidental camera shift. For color matching, the use of a 16 ? 16 histogram with variable bin widths was found to give better performance than uniform histograms. Its performance was comparable to a 5-component Mixture of Gaussians and had a significant advantage in computational speed. To enable the tracker to cope with events such as occlusion, group formation and splits, a heuristics based match assignment algorithm was designed, implemented and was tested to effective. The implemented tracker works at around 12fps.