Projective ICP Visual Odometry

Description

This project implements a projective ICP based visual odometry to estimate both the trajectory of a robot and the 3D map of the visual features.

Explanation and visual/numerical results

The explanation and the visual and numerical results are published on Github Pages - ValerioSpagnoli/VisualOdometry.

Outputs

In the folder outputs there are the results of this project:

• outputs/final_results contains
  • 3D_plots.html: visual results;
  • errors.png: the plot of the rotation and translation errors and translation ratio;
  • estimated_trajectory.dat: id, x, y, z values
  • estimated_world_points.dat: id, x, y, z, appearance_1, ..., appearance_10
  • errors.dat: rotation_error, rotation_ratio, translation_error, translation_ratio.
• In outputs/frame_xxx there are the plots of the errors (mean of chi inliers of that frame), the number of inliers, the kernel thresholds and the dumping factors of that frame (iteration by iteration).

Usage

1. Install the requirements:

   pip3 install -r requirements.txt
   

2. Run the main

   python3 main.py