Fast recognition and precise localization of humanoid soccer robot vision system

Abstract

As the vision demands of humanoid soccer robot,an object recognition approach that concurrently segmenting regions and extracting features was proposed,and a monocular localization method based on the kinematic model of humanoid robot was presented.In the object recognition approach,8-connected neighboring method was adopted to track the boundary of color region;meanwhile the features for recognition were extracted geometrically from the boundary.Then the radial symmetry detector was employed to improve the recognition accuracy.Compared to normal methods,the efficiency of this approach was doubled.The localization method combined the pin-hole camera model and three-links robot model to precisely locate the object in the field of view.The effects of the proposed methods were experimentally validated on a real humanoid soccer robot.