Hello, this is my repo for learning slam

Branch 0.1:

Basic Class for a visual odometry, including class Camera, class Frame, class VO. Eigen, Sophus, cv::Mat is used to express the tramsform of cooridinates. cv::orb, cv::BFM is used to do feature extraction and descriptor matching.

Branch 0.2:

Feature based Method added. Add the function of bundle adjustment of PnP generated pose in G2O. Extend a class in G2O and construct the vertices and edges. Replace cv::BFM with cv:flann.

Branch 0.3:

Add local map for visual odometry. Avoid simple pair-wise matching, instead maintain a map (descriptors cache). Add new struct map & mappoint. Update map with deleting and adding points. Maintain a good map size so that achieve a good speed. Currently 0.01-0.03 s one frame.

Branch 0.4:

Direct (semi-dense) Method added. Direct Semi-dense Method with local map. Fix bugs and update maps from time to time. Currently 0.04-0.09 s one frame. Mainly because of too many points thus over 20,000 edges in G2O. Remain to be optimized.

Video:

Todo: Branch 0.5: Pose graph, PCL, closing loop

#####Count total lines of code: git ls-files | xargs wc -l