Surprise Detection and Visual Homing in Cognitive Technical Systems

Abstract — One important feature of a cognitive system is to interact with its environment and to react in real-time to changes in it. A pure geometric representation of the world is therefore often insufficient. To allow a cognitive system to refer to a prior perception, it has to be able to re-localize itself with respect to its environment. Thus, a vision based agent needs to register the currently seen images within its history. In this paper, we investigate not only an image based surprise trigger, which uses Bayesian probabilistic inference techniques to detect pixel-wise changes, but also an online image based homing algorithm. This is necessary to achieve a location independent surprise trigger. Our algorithms allow a cognitive system to localize itself within its environment and to react online to changes respectively to a prior measurement using only a calibrated camera. Experiments show acceptable results in terms of a robust detection of unexpected changes in the environment. I