Feature-Based Stereo Visual Odometry using Harris, SIFT, and BRISK Feature Detectors

This repo contains a stereo VO implementation from scratch using OpenCV implementations of Harris, SIFT, and BRISK.

A report of the results from the experimentation using the 3 feature detectors can be viewed in 'Feature-Based Visual Odometry Investigation.pdf'.

Getting Started

Environment

1. Python>=3.6
2. Install additional packages

   pip install -r requirements.txt
   

3. If desired to run the code in a docker environment, install docker. Instructions for Linux (Ubuntu): https://docs.docker.com/engine/install/ubuntu/

Prepare Dataset

This repo uses the Canadian Planetary Emulation Terrain Energy-Aware Rover Navigation Dataset available at https://starslab.ca/enav-planetary-dataset/.

1. Set up input and output folders

   mkdir input
   mkdir output
   

2. Download the rover frame transforms file, camera intrinsics file, and human-readable base data folder for Run #1 (note: this folder is 16.8 GB)

   cd ./input
   wget ftp://128.100.201.179/2019-enav-planetary/rover_transforms.txt
   wget ftp://128.100.201.179/2019-enav-planetary/cameras_intrinsics.txt
   wget ftp://128.100.201.179/2019-enav-planetary/run_1/new_data/run1_base_hr.tar.gz
   

3. Extract data folder

   tar -xzf run1_base_hr.tar.gz
   

Run Visual Odometry

The default settings run the visual odometry on the first ~80 seconds of Run #1. All outputs will be automatically saved to the output folder.

1. To run inside a docker container, run the following command in the main directory of the repo:

   docker-compose up
   

2. Instead, to run manually, run the following command in the main directory of the repo:

   python3 ./src/rob501_project.py --input_dir=./input --output_dir=./output