Object recognition, i. e. classification of objects into one of several known object classes, generally is a difficult task. In this paper we address the problem of detecting and classifying moving objects in image sequences from traffic scenes recorded with a static camera. In the first step, a statistical, illumination invariant motion detection algorithm is used to produce binary masks of the scene-changes. Next, Fourier descriptors of the shapes from the refined masks are computed and used as feature vectors describing the different objects in the scene. Finally, a feed-forward neural net is used to distinguish between humans, vehicles, and background clutters.