The common practice in the MPEG video traffic modelling 
is to split video sequence into I, P and B subsequences and to
model the traffic (generated by each subsequence)
separately. This frame-oriented decomposition both allows
avoiding non-homogeneity of modelling data and lead to
linearisation of MPEG video traffic characteristics.
In this paper, the low-complexity regression model of an
MPEG-4 video traffic is proposed. The model combines
regression analysis and a new type of linearised
decomposition, based on partitioning of the phase space into
sub-regions. The model was tested as traffic predictor (for
network management purposes) and traffic generator (for
simulation purposes). Experimental results show good
accuracy of the proposed modelling scheme.