The purpose of this study is to generate the map to estimate the robot position by using stereo cameras and odometry. When creating a map image of the feature points is calculated using the Speeded Up Robust Feature(SURF). Matching of feature points obtained from the cameras using Fast Library for Approximate Nearest Neighbors(FLANN). Accuracy of the map is raised by using the Distance-Based Outlier. And processing time of image processing can significantly reduce the by the use of Graphics Processing Unit(GPU). Self-localization experiments were performed using the map. The robot position and orientation can be estimated with the stereo cameras and odometry.
