In active vision, the configuration of a camera system is adapted automatically in order to acquire the most informative observations for a given task. This paper gives an overview of a probabilistic approach for active vision. Planning the camera parameters is described as a partially observable Markov decision process. Here, the relevant components of the image acquisition task, i.e., object and camera, are represented by means of probabilistic models for incorporating uncertainties into planning. Additionally including reinforcement learning allows a priori unknown probabilistic models.
展开▼
