Conclusion

In connection with the general theories of multimodal integration (see section C.1.1 ), the DI and SI models fall into the first category of models in which fusion occurs at the symbolic level. The RD model falls obviously into the second category since the visual modality is recoded into the auditory modality (which is assumed to be dominant). Finally, the MR model falls into the third category of neo-Gibsonian models.

In connection with the principles of sensors integration proposed by Crowley and Demazeau [75], it can be noticed that the first three principles (common lexicon and coordinates system) cannot be avoided. However, the fourth and the fifth principle (uncertainty and confidence) are seldom considered, even though they are somehow in the essence of Massaro's FLMP. The four models are schematically represented in figure C.5 .

  
Figure C.5 : Schematic of the four integration models

Several variations of these models could be developed. Continuous links from a model to the other could even be elaborated, as it is the case in the modeling area, so that differences would be smoothed at the same time as the structures would become more complex. On the opposite, and in a spirit of methodological control, we will later on fix the architectures in their ``canonical'' form, so that the different options can be clearly compared.

Esprit Project 8579/MIAMI (Schomaker et al., '95)
Thu May 18 16:00:17 MET DST 1995