Upward monocular camera based SLAM using corner and door features

Small-size robots usually employ cheap sensors for navigation instead of expensive laser scanners or stereo cameras. This paper deals with the SLAM process using a monocular camera which heads upward to see the ceiling and the upper portion of a wall. This upward camera has some advantages of being free of dynamic obstacles, the fixed distance to the ceiling and so on. Most past research based on an upward camera used corner features for localization, which are not always extracted in an indoor environment. In this research, however, door features are added to overcome this difficulty involved in SLAM using the corner features only. A door helps not only to estimate the pose of a robot, but also to divide the environment into several meaningful areas. A particle filter is adopted to estimate the door position to check whether the specific door is suitable for the SLAM process before registering it in the EKF algorithm. Experimental results show that the proposed scheme works successfully in various indoor environments.