In [1],[2] we have introduced a new approach for the spatiotemporal segmentation of image sequences. Here a 2D+t sequence is considered as a 3D image, and 2D objects moving in time (or following a given motion model) are segmented as 3D objects with the use of connected morphological filters, and are represented as spatio-temporal flat zones. However when an object undergoes occlusion by another in the sequence, their 3D trajectories intersect, and the spatio-temporal segmentation will fuse the two objects into a single flat zone. In this paper we introduce a method for separating occluded objects in spatio-temporal segmentation. It is based on a study of the changes of topology of the temporal sections of a flat zone. A topologically constrained watershed algorithm allows to separate the objects involved in the occlusion.