Real-Time Dense Visual Odometry for RGB-D Cameras

We propose a visual-inertial localization and mapping (SLAM) algorithm that is lightweight and robust by using an RGB-D camera. It can achieve real-time six-degree-of-freedom pose estimation and dense mapping. In this algorithm, recognizing and filtering dynamic features is added, which will reduce the influence that moving objects incorporate and render pose evaluation more robust. In addition, this approach exploits batch point-to-point and point-to-plane constraints to refine the motion evaluation results, and restrain the cumulative error. We validated the performance of this algorithm on datasets and real-world experiments against other state-of-the-art methods. Finally, we applied the algorithm to an aerial vehicle in GPS-denied environments to accomplish the specified mission and verified the reliability of the algorithm in scenarios with vibration.