Even though many of today's vision algorithms are very sucessfull,
they lack robustness since they are typically limited to a particular
situation. In this talk we argue that the principles of sensor and
model integration can increase the robustness of today's computer
vision algorithms substantially. In this talk we discuss two examples
namely face tracking and face detection where the robustness of simple
models is leveraged by sensor and model integration. The first example
is multi-cue tracking of faces including the principles of
self-organization of the integration mechanism and self-adaptation of
the cue models during tracking. The second example shows how the
maximization of mutual information can be used to combine object
models without prior learning. The same principle can be used also for
model selection.