A crappy visual SLAM implementation I'm working on to learn stuff. Still a work in progress.

refactor coming at some point

video playlist

• add keyframes
• refactor everything
• figure out what bundle adjustment is

• not sure if I can share the datasets i'm using, so just put any video file in the vids/ folder

• currently only have visual odometry barely working, so just run python stream.py to see it working

• check out Multiple View Geometry by Hartley & Zissermann
• Avi Singh's visual odometry projects
• twitchslam by geohot
• UZH and ETH Zurich's course (Vision Algorithms for Mobile Robotics)
• S-PTAM, PTAM, ORB-SLAM, LSD-SLAM
  • papers and code are super useful
• OpenCV references in API documentation
  • look in the SFM modules

• OpenCV
• Numpy
• Scikit-image
• pangolin (we use uoip's fork for its bindings, since its a bit more python-friendly and intuitive)
  • https://github.com/uoip/pangolin

MIT license. Do what you want with it, but please attribute my work if you do. Also don't sue me.