Navid Hasanzadeh

Commodity Wi-Fi devices enable human activity recognition (HAR) without cameras or wearable sensors, offering a non-intrusive, privacy-preserving, and cost-effective solution for smart homes, human–computer interaction, and healthcare. These systems analyze how human motion alters wireless signals captured as Channel State Information (CSI). However, existing Wi-Fi-based HAR methods often struggle to generalize, suffering major performance drops when tested on new users or unseen environments. To address this, we have proposed several methods, including Doppler Radiance Fields (DoRF), a robust motion-representation framework that interprets movement as if observed by multiple virtual one-dimensional Doppler “cameras” around the user. By extracting Doppler velocity projections from CSI and fusing them—analogous to how Neural Radiance Fields (NeRF) reconstruct 3D structure—DoRF forms a latent 3D motion representation. This approach overcomes key limitations of prior CSI-based methods and significantly improves generalization and recognition performance across diverse environments and users (Learn more).

Publications:

1- DoRF: Doppler Radiance Fields for Robust Human Activity Recognition Using Wi-Fi

2- MORIC: CSI Delay-Doppler Decomposition for Robust Wi-Fi-based Human Activity Recognition

3- Hand Movement Velocity Estimation From Wi-Fi Channel State Information

4- Fresnel Zone-Based Voting With Capsule Networks for Human Activity Recognition From Channel State Information (code)

5- Contrastive Representation of Channel State Information for Human Body Orientation Recognition in Interaction with Machines

6- Joint Human Orientation-Activity Recognition Using Wi-Fi Signals for Human-Machine Interaction

7- Dual-Path Model With Fresnel Zone-Based Voting For Human Activity Recognition Using Wi-Fi

Hypertension, commonly known as high blood pressure, poses a major health risk as it often progresses silently, damaging internal organs without noticeable symptoms. Recognizing the need for a user-friendly system capable of precise and continuous blood pressure monitoring without the inconvenience of cuff-based methods, this research aimed to develop a blood pressure estimation and hypertension detection framework with strong generalization to unseen users. In our recent study, we proposed a method that detects hypertension from photoplethysmography (PPG) signals, achieving a generalization sensitivity of 69.1% and an F1-score of 71.6%. Furthermore, we developed a prototype smart phone application that estimates both systolic and diastolic blood pressure by analyzing electrocardiogram (ECG) and PPG signals received via USB or Bluetooth. The application extracts key features such as Pulse Arrival Time (PAT) and leverages pre-trained machine learning models to provide blood pressure predictions.

Publications:

1- Hypertension Detection From High-Dimensional Representation of Photoplethysmogram Signals (code)

2- Blood Pressure Estimation Using Photoplethysmogram Signal and Its Morphological Features

3- Multi-Observation Hidden Semi-Markov Model for Photoplethysmogram Signal Semantic Segmentation