HyphaROS VO Tutorial

Abstract

A simple tutorial for monocular visual odometry implementations.

• python3 with opencv3
• Low dependencies (only numpy, cv2, sys and matplotlib)
• 2D-2D VO pipeline (KLT tracker)
• 3D-2D Localization (PnP RANSAC)

Environment Setup

Recommend to setup a new virtualenv first.

• $ mkvirtualenv hypharos-vo --python=python3
• $ workon hypharos-vo
• $ cd WORKSPACE_PATH/hypharos_vo_tutorial
• $ pip install -r requirements.txt

Download Kitti Dataset

The example in this package uses the 00 sequence of Kitti Odometry Dataset

Operation

Remember to active 'hypharos-vo' virtualenv first:
$ workon hypharos-vo

Visualize all image

$ python visualize_all_images.py -inputs_dir ../../dataset/kitti/00/image_0/ -image_endswith png

Monocular VO (2D-2D KLT)

• Command argument help function:
  python mono_vo_kitti.py -h

• For virtualbox installed version (default path to dataset):
  Simple test: $ python mono_vo_kitti.py
  Verbose Mode: $ python mono_vo_kitti.py --v
  Use GPS scale info: $ python mono_vo_kitti.py --a

• For specific dataset path:
  $ python mono_vo_kitti.py -pose_path ../../dataset/kitti/poses/00.txt -image_dir ../../dataset/kitti/00/image_0/ -image_end png

• To stop during frame processing loop, press 'ESC' on image

Monocular Localization (3D-2D PnP-RANSAC)

• Go to 'pnp' folder and execute the script:
  $ python pnp_localization.py

Developer

• HaoChih, Lin (hypharos@gmail.com)

License

Apache 2.0