FSD-SLAM: a fast semi-direct SLAM algorithm

Abstract Current visual-based simultaneous localization and mapping(SLAM) system suffers from feature loss caused by fast motion unstructured scene in complex environments. Addressing this problem, a semi-direct SLAM algorithm is proposed paper. The main idea of method to combine the point with direct order improve robustness environment scarce visual features low texture. First, enhancement module based on subgraph developed extract image points more stably. Second, an apparent shape weighted fusion for camera pose estimation, which can still work robustly absence points. Third, incremental dynamic covariance scaling studied optimizing error estimation. Finally, optimized pose, face element model designed estimate fuse cloud obtain ideal three-dimensional map. has been tested extensively benchmark TUM dataset real environment. results show that better performance than existing algorithms.