The reconstruction method presented here makes it possible to recover the global surface shape from image sequences. The algorithm has been tested on real data and shown accurate and efficient results on simple objects (i.e., surfaces are at least C2 and they are not locally plane). For more complex objects, occluding contours may not be sufficient to allow a complete perception of the object. This is the case, for example, if the object presents concavities. In addition discontinuities may appear on the observed rims and thus a local approximation of the surface by a paraboloïd is not always adapted. See, for example, a composed object such as a coffee cup with a handle; epipolar correspondence may lead to points which belongs to the cup and to the handle, which are two different surfaces. This shows that topological properties of the observed object must be taken into account in the reconstruction process.