mmWave-radar-signal-processing-and-microDoppler-classification

This is a repository for codes and template data of paper "Experiments with mmWave Automotive Radar Test-bed"

Please cite our paper with below bibtex if you find this repository useful.

@INPROCEEDINGS{9048939,  author={Gao, Xiangyu and Xing, Guanbin and Roy, Sumit and Liu, Hui}, 
booktitle={2019 53rd Asilomar Conference on Signals, Systems, and Computers}, 
title={Experiments with mmWave Automotive Radar Test-bed}, 
year={2019},  volume={},  number={},  pages={1-6},  doi={10.1109/IEEECONF44664.2019.9048939}}

Update

(June 18, 2022) A script for reading binary file has been created read_bin.m.

(June 9, 2022) One raw binary file named '2019_04_30_pbms002_raw_900fs.zip' can be downloaded here for your use. It contains 900 frames with the same radar configuration as the data used below.

(Nov 22, 2021) NEW!!! The micro-Dooler classification part has been updated

Contact

Any questions or suggestions are welcome!

Xiangyu Gao xygao@uw.edu

Software requirement

MATLAB, Python 3.6, Tensorflow 2.0, Jupyter Notebook

Run codes for generating range-angle maps, range-Doppler maps, and 3D point clouds

1. Customize your testbed/FMCW parameter in script:

   ./config/get_params_value.m
   

2. Select the input data ('pms1000_30fs.mat', 'bms1000_30fs.mat' or 'cms1000_30fs.mat') in script:

   generate_ra_3dfft.m
   

3. Run 'generate_ra_3dfft.m' script to get results. For example, the range-angle image, range-Doppler image, and detected 3D point clouds for input data 'pms1000_30fs' are shown below:

4. You can manipulate the algorithm parameters of below commands in "./utils/cfar_RV.m" script to obtain the desired point-cloud results:

   x_detected = cfar_ca1D_square(Dopdata_sum(rani,:), 4, 7, Pfa, 0, 0.7);
   y_detected = cfar_ca1D_square(Dopdata_sum(:, C(1,dopi)), 4, 8, Pfa, 0, 0.7);
   

5. It is optional to stop the range-Doppler image by setting below flag in codes to 0:

   Is_plot_rangeDop = 1;
   

Run codes for generating micro-Doppler maps

1. Customize your testbed/FMCW parameter in script:

   ./config/get_params_value.m
   

2. Select the input data ('pms1000_30fs.mat', 'bms1000_30fs.mat' or 'cms1000_30fs.mat') in script:

   generate_microdoppler_stft.m
   

3. Run 'generate_microdoppler_stft.m' script to get results. For example, the micro-Doppler map for input data 'pms1000_30fs' is shown below:

   

4. You can manipulate the algorithm parameters in "generate_microdoppler_stft.m" to customize the micro-Doppler map properties:

   M = 16; % number of frames for generating micro-Doppler image
   Lr = 11; % length of cropped region along range
   La = 5; % length of cropped region along angle
   Ang_seq = [2,5,8,11,14]; % dialated angle bin index for cropping
   WINDOW =  255; % STFT parameters
   NOVEPLAP = 240; % STFT parameters
   

Train and Test VGG16 classifier

1. Download the template training data, testing data, and trained model from the Google Drive with below link:

   https://drive.google.com/drive/folders/1CC3nluGDral__geL6zIzCK2t5Jrhfwex?usp=sharing
   

   Note that we select part of our training and testing set for your use here and the model was trainied with whole complete training set.

   You may use the above algorithm "generate_microdoppler_stft.m" to create your own training and testing set (micro-Doopler images).

2. Put the decompressed training data, testing data, and trained model in "template data" folder as follow:

   '.\template data\train_data_part'
   '.\template data\test_data_part'
   '.\template data\trained_model\new_epoch10'
   

3. Run the training:

   train_classify.ipynb
   

4. Run the testing

   test_classify.ipynb
   

License

mmWave-radar-signal-processing-and-microDoppler-classification is released under MIT license (see LICENSE).

Acknowlegement

This project is not possible without multiple great opensourced codebase and dataset. We list some notable examples below.

• raw_ADC_radar_dataset_for_automotive_object_detection
• raw_2D_MIMO_radar_dataset_for_carry_object_detection