In smart homes, many context attributes, i.e. small pieces of context, have to be computed from sensors in order to provide adapted services. We observe that many of those context attributes can be deduced from others. This paper presents how methods to propagate belief functions from one frame of discernment to another can be used to compute contexts. It enables the creation of evidential networks, preventing from recomputing many times the same evidence and enabling the construction of higher levels of abstraction often difficult to obtain directly from sensors. The proposed method is based on an intuitive way of thinking evidential networks considering mappings between frames of discernment. An example where this method is essential is presented.