Multi Person Activity Recognition using CSI (Wi-Fi Sensing)

This project investigated two deep learning architectures: CSITime (Yadav et al., 2022) as well as a compact, smaller version of XCeption (Chollet, 2016). Both models were trained and tested on 7 interactions from the HHI dataset (Alazrai et al., 2022). More informations about the dataset is contained in its own subsection.

The project had two aims:

1. To investigate the efficacy of the CSITime architecture on multi-person Wi-Fi sensing.
2. To propose a lightweight model for the multi classification problem.

Project Structure

• models.ipynb: Contains the code for running and testing different models as well as evaluating them.
• Preprocessing.ipynb: Contains code for accessing the raw CSI recordings and preprocessing them.
• TestingArea.ipynb: Contains code for plotting the subcarriers for a single interaction.

This project used:

• NumPy
• matplotlib
• Pandas
• Tensorflow
• sci-kit learn.

Pre-processing

Model Testing and Evaluation