tempEgo3D 0.1.0

A package for estimating 3D ego-motion velocity using millimetre-wave (mmWave) radar.

tempEgo3D

tempEgo3D is a Python package for calculating the 3D ego-motion velocity of a sensing platform from a millimetre-wave (mmWave) radar data. This package implements the proposed ego-motion estimation methods detailed in [Temporally Constrained Instantaneous Ego-Motion Estimation using 4D Doppler Radar](link to be added).

The two methods are:

• TEMPSAC: TEMporal SAmpling Consensus implements a weighted sliding window of sensor measurements biasing the algorithm to estimate the velocity using newer samples in time.
• TWLSQ: Temporally Weighted Least Squares implements a sliding window and temporal weighting scheme for model fitting.

Prerequisites

A method of supplying this package with mmWave data, either using a physical sensor or a pre-existing dataset.

For use with the Coloradar dataset please install the desired dataset from here. For out of the box usage of this package with the Coloradar dataset follow the directory structure implemented here. If you do not need the extra functionality of the coloradar_package simply clone the repository to get the correct directory structure.

Installation

Use the package manager pip to install tempEgo3D.

pip install tempEgo3D

or

Clone the Github Repository

Prereq: Git installation

git clone https://github.com/samuelLovett/tempEgo3D.git
cd tempEgo3D
pip install .

Usage

The main.py script provides an example of how to use each velocity estimation method.

Hyperparameters:

Shared:

• n: Minimum number of samples to estimate parameters
• k: Maximum iterations allowed
• epsilon: Threshold value to determine if a sample is an inlier
• rho: Number of inlier samples required to assert model fits well

TEMPSAC & TWLSQ:

• buffer_size: Size of the sliding window (m)
• fff_lambda: fixed forgetting factor as defined in (14) of [Temporally Constrained Instantaneous Ego-Motion Estimation using 4D Doppler Radar](link to be added). 0 uses only the most recent samples and 1 uses all samples equally.

To change the values presented in [Temporally Constrained Instantaneous Ego-Motion Estimation using 4D Doppler Radar](link to be added), change their definition within the set_kb() , set_tempsac() , set_twlsq() methods.

Results

Comparison of average absolute pose error for different trajectories. The average root-mean-squared absolute pose error is presented in metres.

[| Algorithm | Edgar Classroom Dataset | Intelligent Robotic Lab Dataset | Edgar Army tunnel Dataset | | ------------- |:-------------:| ------------- |:-------------:| | TEMPSAC | 2.812 | 2.352 | 5.615 | | TWLSQ | 2.338 | 2.166 | 5.653 | | KB | 3.644 | 3.603 | 5.672 |

The average absolute pose error for 100 trials is shown in the above Table. Both our methods outperform KB, with TEMPSAC showing an average improvement of 19.5% and TWLSQ showing an average improvement of 25.3%.]: #

A Note on Potential Errors

If you get an error along the lines of

File "C:\path-to-package\tempEgo\dataset_loaders.py", line *some number*, in *some function*
    t, r = read_tf_file(tf_filename)
TypeError: cannot unpack non-iterable NoneType object

It is likely being caused by a path error to a file within the dataset (like the config file or the path to the dataset itself).

License

MIT

Acknowledgement

I would like to acknowledge the support of the bioMechatronics Lab as well as my supervisors Dr. Carlos Rossa and Dr. Sreeraman Rajan in the creation of [Temporally Constrained Instantaneous Ego-Motion Estimation using 4D Doppler Radar](link to be added).

Citation

When citing our work please use:

Citation to be added