Apart from obvious applications in human computer interfaces, a behaviour model could also be applied to various other fields.
A behaviour model could be used to animate crowds in video games. Each face could be synthesised using a model of facial behaviour. Typical motions could be achieved by learning the model on typical videos from the target game. Once the model has learned the visual motions, new videos could be generated stochastically for each avatar present in the crowd. A video of a crowd could then be created by superimposing the generated video sequence for each face.
Moreover, a behaviour model is not limited to rendering. Medical diagnosis is another possible application of such a model. One could for instance model a heart valve from an echocardiogram video sequence. Informations such as the shape, the texture and the temporal behaviour of the heart are contained within that model. A reference model could be created from an echocardiogram video sequence of a healthy person. For each new patient to diagnose, a model of their heart is created and compared with the reference model. The comparison measure derived in this thesis could be used for that purpose. If the result is above a predetermined threshold, the patient is considered having heart problems and therefore in need of further examinations to determine the exact nature of his problem. One could also imagine comparing the model with models of typical heart diseases to determine the most probable disease of the patient.
This application naturally extends the active appearance framework in which an appearance model is used to find instances of an object in an image. A behaviour model could be used to find instances of a particular behaviour of an object in a video sequence.