Research On Real-Time Visual Localizaton For Indoor Exprimental Multiple-Robot System

Mobile robot research has become a hot field in the robot technology. In the course of operation, mobile robots must simultaneously obtain its location and pose. Compared with traditional sensors, visual sensors merit the advantage of wide detecting scope and full information, and thus localization with visual sensors have become one of the main research areas in robot localization.In this paper, a vision system has been designed, which can calculate robots'location and pose rapidly and accurately. To distinguish the robots, polygons are pasted on the ceiling of the robots. In this vision system, common CCD cameras are mounted on the ceiling. Based on certain state, a global or recursive search of the poly is performed. As the search succeeds, sub-pixel accuracy corner detection algorithm is utilized to localize the features more precisely. In the case of global search of the mark, we use a two-step method, which is first to search the area of the robot and then search the mark inside the area of the robot. In the case of recursive search of the mark, Kaman Filter is employed to predict the vertex of the mark and Harris corner detector is then used to search the vertex in the neighboring area.Then, we first localize the detected poly to a coarse accuracy and then use this coarsely reconstructed poly to compare with the designed poly and get the correspondence of the image point and the physical point coordinate in the poly frame. Using vision algorithms, we can calculate the robot's location and pose.As our camera is a common CCD camera and not omni-directional camera, vision field is very limited. To overcome this disadvantage, we use two cameras mounted on the ceiling. The two cameras have common regions. When a robot moves from vision field of a camera to the other camera's vision field, our vision system automatically uses the other camera to localize the robot.In this paper, the performance of the vision system is also tested and analyzed. We conclude that the system has a wide vision field, operates rapidly and accurately. What's more, this system is robust to illumination change and is easy to expand.