One of the central problems in computer vision is the detection ofsemantically important objects and the estimation of their pose. Most of thework in object detection has been based on single image processing and itsperformance is limited by occlusions and ambiguity in appearance and geometry.This paper proposes an active approach to object detection by controlling thepoint of view of a mobile depth camera. When an initial static detection phaseidentifies an object of interest, several hypotheses are made about its classand orientation. The sensor then plans a sequence of views, which balances theamount of energy used to move with the chance of identifying the correcthypothesis. We formulate an active hypothesis testing problem, which includessensor mobility, and solve it using a point-based approximate POMDP algorithm.The validity of our approach is verified through simulation and real-worldexperiments with the PR2 robot. The results suggest that our approachoutperforms the widely-used greedy view point selection and provides asignificant improvement over static object detection.
展开▼
