Entropy-based gaze planning

摘要: Abstract This paper describes an algorithm for recognizing known objects in unstructured environment (e.g. landmarks) from measurements acquired with a single monochrome television camera mounted on mobile observer. The approach is based the concept of entropy map , which used to guide observer along optimal trajectory that minimizes ambiguity recognition as well amount data must be gathered. Recognition itself optical flow signatures result motion — are inherently ambiguous due confounding motion, structure and imaging parameters. We show how gaze planning partially alleviates this problem by generating trajectories maximize discriminability. A sequential Bayes handle remaining accumulating evidence different object hypotheses over time until clear assertion can made. Results experimental system using gantry-mounted presented effectiveness large class common objects.