Abstract/Overview:

This thesis will explore the use of deep learning models to perform both temporal and spatial segmentation of RF signals in order to enhance human activity recognition and indoor localization. The research will investigate how different neural network architectures can be applied to raw or preprocessed RF data (e.g., from Wi-Fi, mmWave) to segment distinct activities and localize individuals within complex indoor environments. The thesis will aim to improve segmentation accuracy by comparing traditional signal processing techniques with state-of-the-art deep learning methods, as well as exploring the potential of hybrid approaches. The outcomes of this research could be applicable in smart homes, healthcare monitoring, and security systems.

Key Step:

  1. Dataset Preparation: Gathering RF sensing datasets (or creating a new one) with labeled human activities and spatial locations.

  2. Preprocessing and Feature Extraction: Signal filtering, noise reduction, and extraction of key features like signal strength, phase, or frequency characteristics.

  3. Segmentation Models:

    • Temporal Segmentation to identify time intervals corresponding to different actions.
    • Spatial Segmentation to partition indoor spaces based on signal reflections or other RF characteristics.

  4. Performance Evaluation: Comparison of traditional methods (e.g., thresholding or clustering) with deep learning-based approaches in terms of accuracy, precision, and real-time performance.

  5. Applications and Use Cases: Testing the developed models in real-world scenarios like human activity detection (e.g., walking, sitting, falling) and indoor localization in environments like smart homes or hospitals.