Keyframe-based RGB-D SLAM for Mobile Robots with Visual Odometry in Indoor Environments using Graph Optimization

The SLAM problem is currently one of the most important topics in mobile robotics, due to the high number of applications that need its solution. This work proposes a methodology to perform SLAM in indoor environments with RGB-D data. The robot motion is estimated using FOVIS, a robust visual odometry system, and a graph-based probabilistic approach is used to minimize the errors caused by the drift in visual odometry. A keyframe selection approach is used to construct the graph and the g2o framework is used for the optimization. The proposed methodology is implemented as a Robot Operating System (ROS) package, and it is evaluated using benchmark datasets available in the literature. A comparison is made with state-of-the-art methods.