抄録
The purpose of this research is to develop a method of generating maps in outdoor environments by a mobile robot with an omnidirectional camera. Maps are effective in guiding robots and people, but making them by hand is a tedious work. We thus adopt SLAM (simultaneous localization and mapping) methods to automatically generate maps. In outdoor environments, since the distance to surrounding objects are much larger than in the case of indoor environments, distance information obtained by, for example, stereo may not be reliable. We thus use a bearing-only SLAM (BO-SLAM) method, which estimates the robot motion and landmark position simultaneously only from the bearings of landmarks. Since the variety of landmark bearings is a key to a reliable estimation, we use an omnidirectional camera system which consists of five conventional cameras. We show the effectiveness of omnidirectional information by simulation and apply the method to real images.
