The layer model describes images as the projection of a three dimensional scene composed of several opaque objects located at different depths from the camera. The background is the last layer, and the moving objects are projected upon it, in opposite order of their depth. When a layer is projected upon another, the one below is hidden, thus creating an occlusion. In the case of a video sequence, the layers move during the sequence and a layer might be hidden by another one during several images. In this article we present a variational method to extract layers from a video, recover the depth order and interpolate layers even in the case of total occlusion. The method is based on a known background. From a uniform motion hypothesis on the 3d objects, a deformation model of the projection on the image is deduced. A variational formulation is used to recover the model parameters, these allow the extraction of moving object layers, depth order and the reconstruction of the sequence