Feature-based, Direct, and Deep Learning Methods of Visual Odometry

Presentation by Yafei Hu, part of the AirLab Summer School 2020. Sessions list, overviews, and links to repos: https://theairlab.org/summer2020 This session is an overview of visual odometry, including feature based, direct and deep learning based methods. Code: https://bitbucket.org/castacks/visual_odometry_tutorial/src/master/ Outline: 0:00 - Pinhole camera projection model 18:37 - Epipolar constraints 26:47 - The essential matrix 30:02 - The fundamental matrix 39:08 - RANSAC algorithm 44:31 - Solve camera pose 44:48 - Feature detector and descriptor 49:14 - Optical flow and LK algorithm 56:03 - Direct methods intro 1:00:32 - ORB-SLAM 1:05:18 - DTAM 1:10:25 - LSD-SLAM 1:14:42 - Supervised learning based VO 1:25:38 - Self-supervised learning based VO 1:28:56 - Hybrid methods 1:32:00 - Exercises Air Lab Website: https://theairlab.org Twitter: https://twitter.com/airlabcmu Facebook: https://www.facebook.com/airlabcmu Medium: https://medium.com/airlabcmu