Choosing the parameters of the MRF model for motion detection

The paper discusses the problem of choosing the parameters of the Markov random field (MRF) model used for motion detection and tracking. When dealing with an arbitrary sequence of images, it is necessary to adjust the parameters of the model to a given sequence. This is a time-consuming operation which cannot be carried out at the time of acquisition of the sequence. The approach proposed by the authors is to use a small number of simple linear models of the brightness of the moving object, for which different sets of parameters are assumed. Each set of parameters allows one to find some portion of the masks of moving objects in real-life images. It was confirmed experimentally that the complete masks of moving objects can then be found by logical summation of the masks obtained for various sets of parameters. An example presented in the paper shows that this approach results in a significant improvement of the masks.