In this paper we incorporate different kinds of information, namely colour, texture, and motion, into a segmentation process. The segmentation is based on a variational framework for vector-valued data using a level set approach and a statistical model to describe the interior and the complement of a region. It is assumed here that there are only two regions in the image, basically one object and the background. In order to obtain appropriate texture features, we apply the idea of the nonlinear structure tensor, which turns out to have very good discrimination properties while inducing only three feature channels. The spatio-temporal version of the nonlinear structure tensor is also used to estimate the optic flow as the feature for motion. Before the actual segmentation process, the features are presmooth- ed by a novel nonlinear diffusion technique closely related to TV flow, but being strictly edge enhancing. The coupling between the channels hereby ensures the enhancement of edges at joint positions in all channels. This kind of presmoothing works well together with the statistical model we use, namely nonparametric Parzen density estimates. We also consider multi-scale segmentation in order to speed up the process and to obtain more robust results.Furthermore, it is shown that our method can not only be applied to segment images but also to track moving objects. The tracking of several moving objects is taken into account, where a coupling between the object regions allows to track also partly occluded objects.Our method has been verified in several experiments using synthetic as well as real image data. There we checked the importance of the different kinds of information for obtaining good results. Finally, also the limitations of our approach are described and shown in some examples.