WiFi-Based Human Action Recognition using Raspberry Pi

Félév: 2023-2024 tavasz

Kategória: Szoftver

Téma leírása

Description

The goal of this student project is to explore and implement a system for human action recognition using WiFi signals captured by Raspberry Pi devices. WiFi signals can be influenced by human movements, and this project aims to leverage this information for recognizing specific actions or gestures. The project involves signal collection, preprocessing, feature extraction, and training machine learning models for action recognition.

Project components

• Background Research: Conduct a literature review on existing methods for human action recognition using WiFi signals. Explore relevant papers, techniques, and applications in the field.
• Raspberry Pi Setup: Set up Raspberry Pi devices with WiFi adapters to capture and record WiFi signals in the surrounding environment. Implement a data collection strategy to capture WiFi signal variations caused by human actions.
• Data Collection: Define a set of human actions or gestures that the system aims to recognize (e.g., walking, sitting, waving). Develop a protocol for collecting WiFi signal data while individuals perform these actions in the Raspberry Pi's proximity.
• Signal Preprocessing: Preprocess the collected WiFi signal data to remove noise, handle missing values, and ensure uniformity in the dataset. Explore and apply techniques for signal denoising and normalization.
• Feature Extraction: Extract relevant features from the preprocessed WiFi signal data that can be used for human action recognition. Investigate time-domain and frequency-domain features, as well as any domain-specific features that may enhance recognition accuracy.
• Machine Learning Models: Train machine learning models (e.g., supervised classifiers) using the preprocessed and feature-extracted data. Explore different algorithms, such as decision trees, support vector machines, or neural networks, and evaluate their performance.
• Model Evaluation: Evaluate the trained models using a separate dataset not seen during training. Measure accuracy, precision, recall, and other relevant metrics to assess the effectiveness of the system.
• Real-time Action Recognition: Implement a real-time action recognition system on the Raspberry Pi, where the WiFi signals are continuously monitored and actions are recognized on-the-fly.
• User Interface (Optional): Develop a simple user interface (web-based or console-based) for users to interact with the system, view recognition results, and provide feedback.

Feltételek

• Programming experience in MATLAB/Python/C++

Külső partner: Nokia Bell Labs

Maximális létszám: 1 fő