A new model for variable bit rate (VER) video traffic is presented. Th
e model, which could be used as a traffic generator, considers two tim
e scales: scenes, for periods of several minutes, and groups of pictur
es (GOP), for periods of half a second. To model the scene changes, a
Markov chain is used. For the GOP level, a modification for the projec
ted autorregresive (PAR) model is proposed so that the fitting of the
autocorrelation function is improved, The model is used to imitate two
real MPEG video signals, showing that it is able to accurately captur
e the behaviour of the real traffic in a queue.