Email

[email protected]

My CV

http://eowjd0512.github.io/

Re-implementation of VITAMIN-E SLAM

This project was started to re-produce the paper

"M. Yokozuka et al, VITAMIN-E: VIsual Tracking And MappINg with Extremely Dense Feature Points, CVPR 2019" https://arxiv.org/pdf/1904.10324.pdf

1. Eigen - the latest version (>= 3.3.9) from https://gitlab.com/libeigen/eigen

2. OpenCV (we tested at the version 4.5.0)

3. Panglin - https://github.com/stevenlovegrove/Pangolin

[Done]

• Architecture construction
• Local extrema feature extraction and curvature tracking
• Initialization (only uses the essential matrix decomposition, not homography, so it doesn't well in a planar scene.)

[TODO]

• Implementation of Subspace Gauss-Newton Method

We used the architecture on the ORB-SLAM System https://github.com/raulmur/ORB_SLAM2. This is because it has a complete and efficient framework suitable for multi-thread-based SLAM.

Thanks to yycho0108 https://github.com/yycho0108/vitamin-e, we can be able to implement the feature tracking part easily.

The code will be further optimized using modern C++ after implementation is complete.