Awesome

Vi-Fi: Associating Moving Subjects across Vision and Wireless Sensors

Full paper (accepted by IPSN'22)

Vi-Fi Dataset

Directory over view:

• data_collection_preprocessing/:

  • scripts of extracting metadata from ZED .svo files
  • scripts of ground truth labeling
  • scripts of data pre-processing and organizing

• Deep_Affinity_Learning/:

  • dataset_v52_better/:

    • scripts of constructing data samples for deep affinity matrix training
    • scripts of spliting training and testing set

  • v52/:

    • scripts of model architecture
    • scripts of training and testing

• Bipartite_Association/src/:

  • MATLAB scripts to prepare and process data of the visual tracker, IMU readings and FTM to associate them.

Execution

Deep Affinity Matrix Learning

Environment & Requirements:

Ubuntu 18.04
CUDA 10.2
matplotlib==3.1.1
numpy==1.19.5
opencv_python==4.0.1.24
pandas==0.25.3
Pillow==8.4.0
pytz==2019.3
scikit_image==0.17.2
scikit_learn==1.0.1
scipy==1.1.0
skimage==0.0
torch==1.8.1
torchvision==0.9.1
torchviz==0.0.1

cd Deep_Affinity_Learning/
pip install requirements.txt
virtualenv -p /usr/bin/python3 [your_env_name]
source [your_env_name]/bin/activate

To train:

cd dataset_v52_better/
python split_data_train_test.py
cd ../v52/
python train_v52.py  [your_dir_to_save_the_model] --fold 1 --epoch [number of epoches] --dataset [your_dir_of_train_test_dataset]/train_test_shuf_split_v2/ --lr [learning rate] --batchSize [batch size]

To test:

python tracklet_ID_assignment_ml_demo.py # need to manually change the directory of .pth in the script

Bipartite Association

Environment & Requirements:

MATLAB 2019a or later
Navigation Toolbox
Curve Fitting Toolbox

To prepare the data:

• run the script: DataPreparation.m
• change src_path and sequences_path to read from your directory.

To associate:

• run ZedMain.m