Stereo Visual Odometry

This project aims to calculate the relative pose between consecutive stereo image pairs by minimizing the reprojection error in 3D-2D motion estimation framework.

Requirements

OpenCV 4.2.0
scipy 
numpy
date_time
system

Project Pipeline

1. Disparity Map Generation
Block matching algorithm has been employed to obtain the disparity map of the scene from the stereo image pair.
2. FAST Feature detection
3. Lucas-Kanade Optical flow

4. Rejection of non-stationary points
Two methods have been implemented and compared to obtain the stationary points in the scene : Inlier detection using the Maximum Clique algorithm and a novel Outlier rejection method which is a variant of the Maximum Clique algorithm that does not rely on an arbitrary threshold as its predecessor.
5. Least squares minimization of the reprojection error involving the 3D and 2D-point correspondences.

Please refer the Stereo_VO_Report.pdf for more details.

Results