Vision Algorithms for Mobile Robotics (VAMR) Mini project

Report can be found here

Team Members

1. Hardik Shah (hashah@ethz.ch)
2. Deepak Ganesh (dganesh@ethz.ch)
3. Aniruddha Sundararajan (asundararaja@ethz.ch)
4. Deepana Ishtaweera (dishtaweera@ethz.ch)

Tested on

Screencasts were recorded on, ROG Zephyrus G15 GA503 GA503QM-HQ121R
OS: Ubuntu 22.04
CPU: 3.0 GHz AMD Ryzen 9 5900HS
RAM: 16 GB 3200MHz

Steps to setup

Setting up conda environment

Note: make sure you have miniconda3/ anaconda installed and working in the terminal
Note: first navigate into the folder

conda env create -f python_env/conda_config.yml

Optional: using pyenv virtual environment

pip3 install -r python_env/requirements.txt

Download the datasets

Use the following commands to download the benchmark datasets.

mkdir data && cd data
wget -O parking.zip https://rpg.ifi.uzh.ch/docs/teaching/2023/parking.zip
unzip parking.zip
wget -O kitti05.zip https://rpg.ifi.uzh.ch/docs/teaching/2023/kitti05.zip
unzip kitti05.zip
wget -O malaga.zip https://rpg.ifi.uzh.ch/docs/teaching/2023/malaga-urban-dataset-extract-07.zip
unzip malaga.zip
mv malaga-urban-dataset-extract-07 malaga
wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=12IQMiJbkg5LW9epJfGxKL8U6VYO33fu7' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=12IQMiJbkg5LW9epJfGxKL8U6VYO33fu7" -O own.zip && rm -rf /tmp/cookies.txt
unzip own.zip

The data folder structure should be as follows.

├── data
│   ├── kitti
│   │   ├──05
│   │   │   ├── image_0
│   │   │   │   ├── ...
│   │   │   ├── image_1
│   │   │   │   ├── ...
│   │   │   ├── calib.txt
│   │   │   ├── times.txt
│   │   ├──poses
│   │   │   ├── ...
│   ├── malaga
│   │   ├── ...
│   ├── parking
│   │   ├──images
│   │   ├──K.txt
│   │   ├──poses.txt
│   ├── own
│   │   ├── ...

Run the app

Usage of the vo_pipeline.py file

usage: vo_pipeline.py [-h] [--dataset_dir DATASET_DIR] [--dataset_name DATASET_NAME] [--config CONFIG]

Visual Odometry Pipeline

optional arguments:
  -h, --help            show this help message and exit
  --dataset_dir DATASET_DIR
                        Path to the dataset directory
  --dataset_name DATASET_NAME
                        Name of the dataset: can be kitti, malaga, parking or own
  --config CONFIG       Path to the config file: can be config/params.yaml, config/params_kitti.yaml, config/params_malaga.yaml,
                        config/params_parking.yaml or config/params_own.yaml

Use the following python commands to run the vo pipeline for different datasets. Make sure to navigate to the folder root before running the commands.

python3 vo_pipeline.py --dataset_name kitti --config config/kitti.yaml
python3 vo_pipeline.py --dataset_name parking --config config/parking.yaml
python3 vo_pipeline.py --dataset_name malaga --config config/malaga.yaml
python3 vo_pipeline.py --dataset_name own --config config/own.yaml

To activate Bundle Adjustment, change the use_ba flag to True in the configs. To activate Bootstrapping when landmarks go below a threshold, change the use_bootstrap flag to True in the configs.

Note: Bootstrapping and Bundle Adjustment cannot be used together.

Note: the results are saved into a subfolder with the dataset name in the out/ folder

Screencasts of the Datasets

Here are the video demonstrations showcasing the monocular visual odometry pipeline. Our approaches have been evaluated using three widely recognized datasets: Parking, Kitti, Malaga, and an additional dataset that we generated ourselves. Visit the playlist on Youtube.

Kitti

Without BA

With BA

Malaga

Without BA

With BA

Parking

Without BA

With BA

Custom Dataset

VO Result

Without BA

With BA

RAW Video

Calibration Video