Visual Odometry Pipeline

Overview

The goal of the project is to develop a monocular Visual Odometry pipeline, incorporating key features such as the initialization of 3D landmarks, keypoint tracking, pose estimation from 2D ↔ 3D correspondences, and the triangulation of new landmarks

Performance

The implemented pipeline demonstrates good quality, exhibiting robustness and effective functionality across various datasets and scenarios. Additionally, our pipeline achieves notable speed, attaining a performance rate of around 30 fps.

1. KITTI Dataset

2. MALAGA Dataset

3. PARKING Dataset

Setup Instructions

1. Download Dataset

Download the dataset from the (VAMR) website.

2. Organize Dataset

Place the downloaded dataset in the data directory at the root of the project.

3. Create Virtual Environment

3.1 Pipenv

To create a virtual environment for the project with Pipenv do the following in the root of this project:

pipenv shell

Install the required libraries from the requirements.txt:

pip install -r requirements.txt

3.2 Anaconda

To create a virtual environment for the project with Anaconda do the following in the root of this project:

conda env create --file=environment.yml

The command above will create the virtual environment and install all the necessary libraries. To activate the environment, type:

conda activate vamr

4. Run the code

Now you can execute the code by typing:

python3 src/main.py

5. Change dataset

To change dataset or any other visualization options, modify the following lines in the main.py file

  # SELECT DATASET
  dataset = Dataset.PARKING
  # IF YOU WANT TO PLOT JUST THE LOCAL TRAJECTORY SET performance_booster = True
  performance_booster = False
  # IF YOU WANT TO COMPARE THE TRAJECTORY WITH THE GROUND TRUTH SET ground_truth_mode = True
  ground_truth_mode = True
  # IF YOU WANT TO PLOT THE FINAL COMPARISON SET final_comparison = True
  # - WARNING: IT WILL NOT PLOT THE TEMPORARY RESULTS
  # - FOR THE MALAGA DATASET NO ground_truth IS PROVIDED
  final_comparison = False