Marker-Less Augmented Reality for Human Robot Interaction

This paper presents the marker-less augmented reality system for in-situ visualization of robot's plans to the human operator. The system finds the natural features in the environment and builds the 3D map of the working space during the mapping phase. The stereo from motion method is utilized to compute the 3D position of natural features, while the position of the camera is computed from the artificial markers placed in the working space. Therefore the map is build in the fixed frame of reference frame provided by artificial markers. When the whole working space is mapped, artificial markers are not required for the functionality of the augmented reality system. The actually seen natural features are compared to those stored in the map and camera pose is estimated according found correspondences. The main advantages are that no artificial markers are necessary during regular use of the system, and that method does not rely on the tracking. Even the single frame is sufficient to compute the pose of camera and visualize the robot's plan. As there is a big number of natural features in the environment, the precision of the camera pose estimation is sufficient, when the camera is looking into the mapped working space.