ORB-SLAM-based tracing and 3D reconstruction for robot using Kinect 2.0

We aim to track robot location and provide dense 3D reconstruction while exploring environment in real time, based on one of the best SLAM algorithm called ORB-SLAM, which accurately estimate the camera poses and sparse 3D map of the environment based on images. However, the sparse map can't be applied to obstacle avoidance and navigation. We base on octrees and use probabilistic occupancy estimation, extending the mapping technique of ORB-SLAM and builds 3D reconstruction which can be used in robot field, the validity of this method is verified via benchmark datasets. Finally, the extended SLAM system is applied to a handheld Kinect 2.0, experiments show that the camera pose is accurately tracked and octomap is completed in real time.