Visual SLAM with oriented landmarks and partial odometry

In this paper, we describe a system that can carry out visual simultaneous localization and mapping (VSLAM) in structured indoor environments; this system uses a small-sized low-cost commercial humanoid robot which uses a camera mounted on a pan-and-tilt device (it was designed and built by us) as his head. Almost all of the VSLAM approaches use points 2D like features of the robot's environment for estimate his pose and the structure of his environment. We propose a new approach to get points 3D (i.e. feature's position and orientation with respect to the current robot's position are known) this approach is based in a monocular 3D reconstruction scheme; with this information we can improve the results in classical SLAM EKF method. Experiments had made in a structured environment, we implemented them for two cases, though; the first one was implemented without odometry and second one was carried out with partial odometry information; that means we included robot's orientation in EKF algorithm. This information had gotten with an IMU (Inertial Measurement Unit), which we installed in the robot. Finally, we want to point out that all the components used into the robotics' platform are low cost.