Implementation and verification for a moving object localization system using trinocular stereo vision

摘要

We have proposed a robot vision system for 6-DOF localization (Translation:3, Rotation:3) for moving objects. Existing object localization systems have been difficult to deal with moving objects without initial positioning. In solve this problem, we integrate a 3-D object recognition task and a 3-D object tracking task through a large amount of frame buffer which absorbs a computational time-lag between the two tasks. The 6-DOF localization is achieved by matching a 3-D geometrical object model to an input stereo image sequence from a set of trinocular stereo cameras. In this report, we describe the current implementations of this system and present experimental results to demonstrate that the system can deal with objects of any shape including free-form objects. In addition, we demonstrate how complex problems can be solved in terms of processing time and discuss the possibility of applying the system to intelligent robots.