Résumé : Moment invariants are important shape descriptors in computer vision. In this paper, we give a general and systematic method to derive moment invariants of any order and for k-dimensional space (invariants to scale, to orthogonal transformation and to translation). This result is applied to the selection of adequate features for visual servoing. To elaborate the control law, we also determine the analytical form of the interaction matrix of any image moment. This matrix relates the variation of the image moments to the 3D motions of the camera. We finally presents some visual servoing experimental results using very simple images. These results allow to validate the theoretical developments proposed in this paper.