roboticsai / python-vo

A simple python implemented frame by frame visual odometry with SuperPoint feature detector and SuperGlue feature matcher.

Python-VO

A simple python implemented frame by frame visual odometry. This project is inspired and based on superpoint-vo and monoVO-python.

We tested handcraft features ORB and SIFT, deep learning based feature SuperPoint, more feature detectors are also possible to be added to this project. For feature matchers, we tested the KNN and FLANN mathers implemented in OpenCV, and the novel deep learning based mather SuperGlue.

Feature detectors

• ORB (OpenCV implementation)
• SIFT (OpenCV implementation)
• SuperPoint

Feature matchers

• KNN, FLANN (OpenCV implementation)
• SuperGlue

SIFT Keypoints

SuperPoint Keypoints

SIFT+FLANN Matches

SuperPoint+FLANN Matches

SuperPoint+SuperGlue Matches

Install

• Get this repository

  git clone https://github.com/Shiaoming/Python-VO.git
  cd Python-VO

• Install python packages

  pip install -r requirements.txt

Run

1. edit dataset path in params/*.yaml;
2. run python main.py --config params/*.yaml in terminal.

For example, to evaluate the SuperPoint with SuperGlue, run:

python main.py --config params/kitti_superpoint_supergluematch.yaml

Evaluations

Absolute and relative translation errors on KITTI sequence 00

Average relative translation errors on KITTI sequence 00

orb_brutematch	sift_flannmatch	superpoint_flannmatch	superpoint_supergluematch
0.748m	0.085m	0.177m	0.103m

Trajectory of ORB feature with brute matcher on KITTI sequence 00

• red: ground truth
• green: estimated trajectory

Trajectory of SIFT feature with FLANN matcher on KITTI sequence 00

• red: ground truth
• green: estimated trajectory

Trajectory of SuperPoint feature with FLANN matcher on KITTI sequence 00

• red: ground truth
• green: estimated trajectory

Trajectory of SuperPoint feature with SuperGlue matcher on KITTI sequence 00

• red: ground truth
• green: estimated trajectory