gorbslam

README

gorbslam

Monocular ORB-SLAM in Go. Based on the paper:

ORB-SLAM2: an Open-Source SLAM System for Monocular, Stereo and RGB-D Cameras
Raul Mur-Artal, Juan D. Tardos
arXiv:1610.06475 [cs.RO]

Currently you can run the following binaries:

• orbcheck: run the ORB feature extractor on an image or on a video capture device.
• parsevocab: parse an ORBvoc.txt file to create an ORB vocabulary.

Requirements

• OpenCV 4+ : brew install opencv or equivalent on Linux.

Status

• ORB feature extration: https://www.youtube.com/watch?v=4AvTMVD9ig0 (describes SIFT but same principles apply)
• Lens distortion correction / camera intrinsics: https://www.youtube.com/watch?v=26nV4oDLiqc
• ORB Vocabulary from file
• Bag of Visual Words to store features for relocalisation (e.g during tracking failure): https://www.youtube.com/watch?v=a4cFONdc6nc
• Bundle adjustment / reprojection error: https://www.youtube.com/watch?v=lmj2Jk5tl60
• Tracker goroutine to localise the camera with every frame (reprojection error and motion-only bundle adjustment): https://www.youtube.com/watch?v=0I30M6yTklo&t=191s
• Local mapping goroutine to manage the local map and do local bundle adjustment: https://youtu.be/0I30M6yTklo?t=281
• Loop closing goroutine using pose-graph optimisation (then full bundle adjustment).

Long-term:

• Stereo ORB-SLAM: fixes scale issues with Monocular and is generally more accurate.