This study is aimed at developing a method of analysis of the 3D structure of a scene considering a monocular image sequence, with an uncalibrated camera -as for an active visual system- and using a continuous model of motion. Surprisingly perhaps, this problem has not been studied much in literature except \cite{vieville-faugeras:95}, but only preliminarly, and without any reference to active vision. This difficulty might have its source in the intrinsic complexity of the underlying equations, which yields a heavy implementation and are thus a-priori not robust. Moreover important developments of analytic equations are not possible as it is the case for calibrated systems \cite{vieville-clergue-etal:95,chaumette-boukir:91,boukir:93}, because of the algebraic complexity of the equations. In order to overcome this difficulty, we have attempted to develop a simplified parameterization of the problem in the case of two or more views, considering a scene with a set of stationary objects and applying an orthographic model of the projection. In this case, fusion along the image sequence is trivial. Thanks to the integration of active visual perception, we demonstrate that it is always possible to generate a displacement so that the previous model is valid, and we can then very easily reconstruct the observed scene. In the case where the motion constraints are approximately verified, we can show that the model is still approximately valid close to the retina. At an experimental level, we report a small implementation taking an image sequence as input, which allows us to compute the retinal motion fields and calculate the reconstruction up to a particular affine transform of the scene.