This is an implementation of the journal "Real-Time Stereo Visual Odometry for Autonomous Ground Vehicles" by Andrew Howard
In A2 step, the author use a rectangular as descriptors, yet we use the opencv built-in ORB method which use BRIEF descriptor In A2 and A3 steps, the author use SAD to calculate a score matrix to match the features. We use the opencv function BFMatch.match() instead.
A5 step which calculate the movement by matches remains unfinished A6,7 steps which evaluate the ORB VO result is unfinised The output of trajectory, which is required to be a .txt file is not finished yet.
The main.py is our work which could be run with RealSense attached. The opencv_viewer and tutorial_depth are two referenced file included in the libRealSense/wrappers/python/examples/...