Skip to content

azinke/rwu-dataset

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

42 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

RWU-Dataset

This tool allows you to access and post-process the recordings of the RWU-Dataset. It only hosts currently recordings of the cascaded RADAR but can be easily extended to other sensors (Camera, Lidar, IMU, etc.).

The Radar dataset contains the raw ADC samples of the cascaded MMWCAS-RF evaluation module of Texas Instruments.

Environment setup

Note: At least python 3.10 is needed to set up this tool for usage.

Create a python virtual environment and install the required packages as follows

python -m venv venv
source ./venv/bin/activate

python -m pip install -r requirements.txt

RWU dataset

TODO: Add the link to download the dataset

Setup and configuration

The structure of this repository is as follows

.
├── core
├── dataset
│   ├── calibration
│   │   └── cascade
│   │       ├── antenna_cfg.json
│   │       ├── coupling_calibration.bin
│   │       ├── coupling_cfg.json
│   │       ├── frequency_calibration.bin
│   │       ├── phase_amp_calibration.bin
│   │       └── waveform_calib_cfg.json
│   └── dataset.json
├── __init__.py
├── rwu.py
├── README.md
└── requirements.txt

The folder calibration/cascade holds the calibration data of the cascaded radar sensor. The recordings must be downloaded and unzipped in the dataset folder. Then, the folder structure would look like this:

.
├── core
├── dataset
│   ├── parking-05092200
│   ├── parking-05092201
│   ├── parking-05092202
│   ├── parking-05092203
│   ├── parking-05092204
│   ├── room-02092200
│   ├── room-05092201
│   └── room-05092202
│   ├── calibration
│   └── dataset.json
├── __init__.py
├── rwu.py
├── README.md
└── requirements.txt

The naming convention used for the folders containing recordings is <name>-<dd-mm-yy><nn>. With <name> a name describing the scene or the place of the recording, <dd-mm-yy> a 2-digit date format (d: day, m: month, y: year) and <nn> a 2-digit order number. Each of those data folders should contain a configuration file (confg.json) that describes the waveform configuration used to record it.

In the snippet above, one can notice that multiple subset of the dataset are present in the dataset folder. Each of those subsets is expected to have the following structure:

.
└── <name>-<dd-mm-yy><nn>
    ├── cascade
    |   ├── adc_samples
    |   |   ├──frame1.bin
    |   |   .
    |   |   .
    |   |   .
    |   |   └──frame<n>.bin
    |   ├── pointclouds
    |   |   ├──radar_pcl1.bin
    |   |   .
    |   |   .
    |   |   .
    |   |   └──radar_pcl<n>.bin
    |   └── config.json
    └── velodyne
        ├──<velodyne frame 1>.cvs
        .
        .
        .
        └──<velodyne frame n>.cvs

The cascade/adc_samples should contain the radar frames (repacked data files from mmwave-repack); And the Velodyne's recordings under the subfolder velodyne. Folder cascade/pointclouds contains already processed point-clouds

The key point to configure the supported subset of the dataset and how to access them is the dataset/dataset.json file.

The only part that requires your attention is the folders key of the datastore entry.

    "folders": [
      {
        "codename": "parking0",
        "path": "parking-05092200"
      },
      {
        "codename": "parking1",
        "path": "parking-05092201"
      },
      {
        "codename": "parking2",
        "path": "parking-05092202"
      },
      {
        "codename": "parking3",
        "path": "parking-05092203"
      },
      {
        "codename": "parking4",
        "path": "parking-05092204"
      },
      {
        "codename": "room0",
        "path": "room-05092200"
      },
      {
        "codename": "room1",
        "path": "room-05092201"
      },
      {
        "codename": "room2",
        "path": "room-05092202"
      }
    ]

In the example shown above, each subset of the dataset to handle, is registered along with a short codename to access it. The subset already mentioned can be accessed with their corresponding codename. New subsets of the dataset can be added similarly. The codenames can even be updated to suit the naming convention that you would prefer. With that, you're all set to play around with the dataset.

Usage

IMPORTANT NOTE:

  • If you've setted up a virtual environment, don't forget to enable it first
  • Rendering are done based on a web based backend. So, a web browser tab will automatically be launched for all renderings.

The easiest way to have an overview of all the available options to interact with the dataset is the help command.

python rwu.py -h

However, find below the cheat-sheet of this CLI tool

  1. Overview
# Print the 'dataset.json' configuration file
python rwu.py -o
python rwu.py --overview

Either one of this command pretty print the entire dataset/dataset.json file to allow a quick overview of the current configuration in use.

Since each subset of the dataset receives a codename to interact with it, you can request the list of currently registered subsets of the dataset and their codenames as follows:

# Get the list of registered dataset and their codenames
python rwu.py -o --codename
python rwu.py --overview --codename
  1. Velodyne sensor
# Render Velodyne lidar 3D pointcloud
python coloradar.py --dataset <codename> -i <frame-index> --velodyne

# Render Velodyne lidar pointcloud bird eye view
python coloradar.py --dataset <codename> -i <frame-index> --velodyne -bev
python coloradar.py --dataset <codename> -i <frame-index> --velodyne --bird-eye-view

See examples below:

# Render Velodyne lidar 3D pointcloud
python coloradar.py --dataset lidar0 -i 130 --velodyne

# Render Velodyne lidar pointcloud bird eye view
python coloradar.py --dataset lidar0 -i 175 --velodyne -bev
  1. Radar sensor

The shorthand used to access the cascaded chip radar data is ccradar. Therefore, we have the following commands

To render already processed point-clouds, the following command can be issued:

# Render radar pointcloud
python rwu.py --dataset <codename> -i <frame-index> --ccradar

# Render cascaded chip radar birds' eye view from 3D pointcloud
# The resolution is in meter/pixel
# 0.1 -> 10cm / pixel
python rwu.py --dataset <codename> -i <frame-index> --ccradar -bev --resolution <resolution>
#
# Processing or raw ADC samples
#

# Render cascaded chip radar  4D heatmap  from raw ADC samples (in polar coordinate)
# --min-range allows to skip very close range pointclouds that are not often coherent
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw --polar [--min-range <range>]

# Render cascaded chip radar  3D heatmap  from raw ADC samples (in cartesian coordinate)
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw [--min-range <range>]

# To render the heatmap with velocity as the fourth dimension, use the `--velocity-view` argument
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw --velocity-view

# The threshold for filtering the heatmap can be adjusted with `--threshold`. The value
# of the threshold is expected to be in the interval [0; 1]
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw --threshold <value>


# Render cascaded chip radar  2D heatmap  from raw ADC samples
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw --heatmap-2d

# Render cascaded chip radar  3D pointcloud  from raw ADC samples
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw --pcl

# Render cascaded chip radar pointcloud bird eye view  from raw ADC samples
python rwu.py --dataset <codename> -i <frame-index> --ccradar --raw --pcl -bev

See examples below:

# Render cascaded radar 3D heatmap skiping the first 1.5m range
python rwu.py --dataset parking1 -i 130 --ccradar --raw --min-range 1.5

# Render cascaded radar 3D heatmap skiping the first 1.5m range, with velocity as the fourth dimension
python rwu.py --dataset parking2 -i 498 --ccradar --raw --polar --min-range 1.5 --velocity-view

# Render cascaded radar 3D heatmap with a threshold value of 0.4
python rwu.py --dataset parking1 -i 65 --ccradar --raw --min-range 1.5 --threshold 0.4

# Render cascaded chip radar pointcloud bird eye view from raw ADC samples
python rwu.py --dataset parking4 -i 175 --ccradar --raw -pcl -bev
  1. Batched processing and save outputs

You can note that the index option -i is no longer needed. The path given for the save-to option could be a non-existing one. The path will automatically be created in that case.

# Render and save all cascaded chip radar plointcloud bird eye view of a given subset of the dataset
python rwu.py --dataset <codename> --ccradar --raw -pcl -bev --save-to <output-directory>

# Render and save all Velodyne lidar plointcloud bird eye view of a given subset of the dataset
python rwu.py --dataset <codename> --velodyne -bev --save-to <output-directory>
  1. Save post-processed files as .csv or .bin files

The placeholder <ext> could be either csv or bin. Binary files are saved as float32 values.

  • csv: Comma Separated Value
  • bin: Binary
# Save all cascaded chip radar plointcloud of a given subset of the dataset as "csv" or "bin" files
python rwu.py --dataset <codename> --ccradar --raw -pcl --save-as <ext> --save-to <output-directory>

# Example for saving post-processed pointcloud as csv files
python rwu.py --dataset parking0 --ccradar --raw -pcl --save-as csv --save-to output

If binary files have been generated, they can be read as follows:

import numpy as np

# [0]: Azimuth
# [1]: Range
# [2]: Elevation
# [3]: Velocity
# [4]: Intensity of reflection in dB or SNR
data = np.fromfile(fileptah, dtype=np.float32, count=-1).reshape(-1, 5)
  1. Animation
# Create a video out of the images present in the input folder provided
python rwu.py --dataset <codename> --animate <path-to-image-folder>

The generated video is saved in the same folder as the images.

Releases

No releases published

Packages

No packages published

Languages