Calculates the Visual Odometry on the KITTI Dataset using sparse features and a combination of RANSAC, 5-Points-Algorithm and Nonlinear Least Square minimisation of the epipolar error. Most important features:
- Mono-camera odometry with Python
- Only the current and the previous frames are used
- Jacobian matrix defined in the state space and not in the tangent space
This approach is not anymore state-of-the-art - rather a nice math excercise. Neverthless, it can be used as a baseline-implementation to measure the performace of other, more modern approaches.
Sparse flow vectors. This approach tolerates a large number of outliers. In white: inlier optical flow vector in the pre-defined road region.
Estimated trajectory and ground-truth.