Contains C++ Opencv implementation of Visual SLAM (IN PROGRESS)
Please note that this is a IN PROGRESS work so there is a lot of death code, the code is not optimized and have unnecessary variables and other stuff only for debugging.
Currently only Visual odometry is implemented:
- I used ORB to extract keypoints.
- Keypoint matching between consecutives frames.
- I used projective geometry constraints to refine matches (Fundamental matrix and homography).
- Once matches are relatively OK I can use camera calibration parameters to find Essential matrix, camera pose and triangulate the keypoints to estimate their 3D position up to scale. During this process a final oulier rejection is performed to keep only few but sure points as landmarks.
TODO:
- Local optimization with g2o or other optimization framework.
- Define a good data structure and representation for mapping (for embedded platforms)
- Global optimization
- Use all this for path planning
