This is a re-implementation of ORB Slam 3 library by Carlos Campos, Richard Elvira, Juan J. Gómez Rodríguez, José M. M. Montiel, Juan D. Tardos..
This video shows the performance of the library on the TUM database.
Currently we have implemented only the monocular case without IMU sensors. We imagine the future roadmap as follows:
- Test the library on more datasets. Discover and fix bugs.
- Use C++14 (or later) features to optimize multithreading efficacy.
- Implement stereo and rgbd cases. (should not take much time)
- Add IMU sensor fusion.
- Test the library in real world scenarios, say, by putting it on a mobile robot or a RC UAV.
The main.cpp program is for testing purposes only and is written to work only with the TUM corridor1 database or a calibrated usb web camera. Implementing a similar application for another database is straightforward.
- The application depends on Eigen3 library for matrix operations.
- For the time being we use OpenCV, however, the dependency is minimal and will be removed in future.
- We use g2o for error minimization problems.