Wandatch: Infrastructure-Free Point-to-Command with Smartwatches and Speakers

This repository contains the official artifact for the paper "Wandatch: Infrastructure-Free Point-to-Command with Smartwatches and Speakers".

Links

• Code: https://github.com/leenchen0/Wandatch.
• Dataset: https://doi.org/10.5281/zenodo.18390782.

Structure

To ensure the code runs correctly, the folders must be organized as follows:

.
├── dataset/
│   └── gesture/
│   └── indoor-localization/
├── exps/
│   └── gesture/
│   └── indoor-localization/
└── README.md

dataset/: This directory contains all the data files required for the experiments.

exps/: This directory contains the source code and scripts files used to run the experiments.

Environment

Matlab:

• Version: R2022b.
• Required Toolboxes (Critical):
  • Signal Processing Toolbox: Required for signal filtering and transformation.
  • Statistics and Machine Learning Toolbox: Required for cdfplot and statistical analysis.

Note for Standalone Execution: If you do not have a MATLAB license, we provide a standalone .exe for the MATLAB (localization and single-stroke recognition). This requires the free MATLAB Runtime (R2022b).

• Download MATLAB Runtime R2022b

Python:

Creating an environment based on Anaconda by running the following commands:

conda create -n wandatch python=3.10
conda activate wandatch

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126
pip install scikit-learn numba fastdtw

Experiments

Indoor Localization

Important: You must set the MATLAB working directory to the folder containing the script (e.g., <root>/exps/indoor-localization/) before running. This ensures relative paths for util/ and dataset/ are resolved correctly.

Fig. 18: Run the script exps/indoor-localization/Fig18.m with Matlab.

Fig. 19: Run the script exps/indoor-localization/Fig19.m with Matlab.

Micro Gesture Recognition

Run the script exps/micro/run.py with Python:

cd exps/micro
conda activate wandatch
python run.py --training

The console will display the accuracy, false alarm rate (FAR), and F1 Score.

Single-Stroke Gesture Recognition

Important: You must set the MATLAB working directory to the folder containing the script (e.g., <root>/exps/single-stroke/) before running. This ensures relative paths for util/ and dataset/ are resolved correctly.

Run the script exps/single-stroke/SingleStroke.m with Matlab. The console will display the average accuracy and the confusion matrix.

Multi-Stroke Gesture Recognition

Run the script exps/multi-stroke/run.py with Python:

cd exps/multi-stroke
conda activate wandatch
python main.py --training