The problem of moving person tracking, without knowledge about the number of persons in the scene, and by taking into account occlusion, under-segementation and over-segmentation, is challenging. A first motion detection gives us regions with several segmentation problems due to bad acquisition conditions. The tracking step, which has to manage all these problems, is realized with the EM algorithm (Expectation Maximization). It uses a kinematic model: we suppose a rectilinear and uniform apparent motion, this hypothesis seems very restrictive but remains locally accurate in most of applications. Good results are obtained with this approach on several sequences, without any initialization.