Robotics-Vision

Implementations and explorations of several pipelines and algorithms in Robotic Vision, as part of the Mobile Robotics (Monsoon 2023) class.

Contents

Transformations-And-Mapping

This set consists implementations of:

1. Euclidean transformations.
2. Ways of representing rotations , interpolation between rotation matrices using SLERP.
3. Waypoint generation and trajectory visualization for the letter M.
4. 3D Mapping from RGBD Data with a specific scene in AI2Thor, creation of a point cloud of the scene and generating Occupancy Grid Maps of the environment from different heights.

ICP-SLAM and Pose Graph Optimization

This set consists implementations of:

1. Non-Linear least squares optimization for Gaussian Function using Levenberg Marquardt (LM) algorithm.
2. Procrustes alignment on two point clouds with (given) known correspondences
3. ICP algorithm with unknown correspondences.
4. Pose Graph Optimization:
   • PGO for 1D SLAM
   • PGO for 2D SLAM
   • Trajectory Evaluation and g2o optimization

Vision

This set consists implementations of:

1. Estimation of fundamental matrix and plotting epipolar lines and epipole given two images of the same scene taken from different view-points.
2. Visual Odometry: recovering the egomotion (the trajectory) of an agent using only the input from the camera or a system of cameras attached to the agent.
   1. Feature extraction using SIFT detector.
   2. Estimation of essential matrix within RANSAC scheme.
   3. Obtaining transformations and plotting both ground-truth and calculated trajectories.
3. Stereo Dense Reconstruction: generating a dense 3D point cloud reconstruction of a scene from stereo images.