The task of obtaining a reliable model of the background of a scene is hampered by many undesirable effects such that sensor noise, camouflage, cast shadows and others. Moreover, the model must be built in real time, as new frames are captured. Stochastic approximation is a methodology to estimate unknown parameters of a dynamic system, i.e. one which evolves through time, under the presence of noise. Consequently, it is suitable for the background modeling section of computer vision systems. Stochastic approximation is based on the online update of the parameter estimates. The updates are done in a way that tend to remove the noise in the observations on the long run. The methods that we consider here are based in the Robbins-Monro stochastic approximation algorithm [10,20, 21, 29, 31].