Direct-ORB-SLAM: Direct Monocular ORB-SLAM

Abstract

Abstract Simultaneous Localization and Mapping (SLAM) plays an important role in the computer vision and robotics field. The well-known ORB-SLAM extracts ORB features and uses feature matching to estimate the camera pose and construct a sparse map. However, the features extracted in the common frame (relative to the keyframe) are not used again except in this camera pose estimation stage, which will cause the waste of computation resources. In this work, we propose a new SLAM framework, i.e. Direct-ORB-SLAM. It uses image intensity for data association in common frames to avoid the redundant computations and performs the relocalization and loop closure detection in ORB-SLAM2 for keyframes. The photometric calibration is also employed to benefit for accuracy and robustness. The popular public dataset EuRoC is adopted to approve the efficiency of proposed method. Results indicate that the proposed Direct-ORB-SLAM algorithm runs two times faster than the well-known ORB-SLAM at the expense of a slight reduction in accuracy.