P³ID: A Privacy-Preserving Person Identification Framework Towards Multi-Environments Based on Transfer Learning

This is the official implementation of the paper: P³ID: A Privacy-Preserving Person Identification Framework Towards Multi-Environments Based on Transfer Learning

About P³ID

We propose a multi-environments person identification framework based on transfer learning using Impulse-Radio Ultra-Wideband (IR-UWB) radar dataset. A neural network is devised for mapping signals from distinct environments into a unified feature space and further align them, enabling the model to extract environment-insensitive features.

Getting Started

Dependencies

pytorch==1.12.1
tensorboard==2.10.1
torchvision==0.13.1
configargparse==1.4
numpy==1.21.5
scikit_learn==1.1.2
timm==0.5.4 

Dataset

Using a real IR-UWB radar testbed, we build a dataset with 22,264 samples from three environments, varying in testing distance and occlusion condition. The directory structure is:

│path/to/dataset/
├──A_train/
│  ├── p1-A-0.5m-F-1
│  │   ├── 1.png
│  │   ├── 1_mw.png
│  │   ├── 1_pt.png
│  │   ├── ......
│  ├── ......
├──A_valid/
│  ├── p1-A-0.5m-F-1
│  │   ├── 2.png
│  │   ├── 3.png
│  │   ├── ......
│  ├── ......
├──A_test/
│  ├── p1-A-0.5m-F-1
│  │   ├── 4.png
│  │   ├── 5.png
│  │   ├── ......
│  ├── ......

<p1-A-0.5m-F-1> means Person 1 conducted the first experiment in Environment A at a distance of 0.5 meters from the radar equipment, with no obstructions.

The anthropometric data of individuals is:

Person ID	Height (cm)	Weight (kg)	Gender
1	182	75	Male
2	180	74	Male
3	175	65	Male
4	168	65	Male
5	170	65	Male
6	175	75	Male
7	186	92	Male
8	170	55	Male
9	162	48	Female
10	160	70	Female

It is worth noting that the initial version of our work provides a part of dataset , while the remaining dataset will be made available in subsequent versions.

Usage

Frist, clone the repository locally:

git clone https://github.com/hxhebit/P3ID.git

Then, install Pytorch, tensorboard, and other dependencies:

pip3 install -r requirements.txt

Next, configure parameters at file1 or file2.

Finally, to train and test on a single node with 8 GPUs, run:

bash run/runs/task1.sh <FOLDER_NAME>

To customize your dataset, simply substitute the dataset with an identical directory structure.

---

This project is continuously being updated.