Motion detection from mobile platforms is a challenging task. It requires precise position information, which is difficult in cluttered dynamic environments. We combine motion detection and position estimation using Expectation Maximization. To reduce the computational time of this iterative approach, we segment the scan into feature elements which are then compared against an a-priori map. The motion detection is tested on real world data from a mass exhibition, showing correct classification. As one application of this information, we present a path planning approach with a unified probabilistic formalism for dynamic and static elements incorporating different levels of knowledge. The resulting probabilistic navigation function derives from co-occurrence probabilities and is currently computed using grids.