A-Simple-Stereo-SLAM-System-with-Deep-Loop-Closing | simple stereo SLAM system with deep loop closing module

A-Simple-Stereo-SLAM-System-with-Deep-Loop-Closing is a C++ library. A-Simple-Stereo-SLAM-System-with-Deep-Loop-Closing has no bugs, it has no vulnerabilities and it has low support. You can download it from GitHub.

The system contains three thread:. In Frontend, it will track the motion based on feature points and LK flow. If the number of tracked keypoints is lower than a thresold, it will detect new features and create a keyframe. Mappoints are created by triangulating the matched feature points in left/right images. In Backend, it will maintain a global map and an local active map. The active map is like a sliding window, containing a fixed number of keyframes and observed mappoints. Optimization of the active map is done in Backend. In LoopClosing, it will first try to detect a Candidate Loop KF of the Current KF using DeepLCD. If succeed, it will then match the keypoints in Candidate KF and Current KF, which is used to compute the correct pose of Current KF using PnP and g2o optimization. If the number of inliers is higher than a threshold, the loop detection will be considered as a success, and loop correction is applyed: first, it will correct the keyframe poses and mappoint positions in active map; second, a pose graph optimization of the global map will be applied. There must be some mistakes in the project as I am just a newcomer to visual SLAM. Please open an issue if you find any problem, and I will be deeply grateful for your correction and advice.