Navid Hasanzadeh

The use of commodity Wi-Fi devices for recognizing human activities eliminates the need for attached sensors and cameras. This non-intrusive, privacy-preserving, and cost-effective method could enhance smart homes, human-computer interaction, and healthcare applications like elderly care and fall detection. Wi-Fi-based HAR systems work by analyzing how human movements affect wireless signals. Channel State Information (CSI) from Wi-Fi devices captures these effects, reflecting how movements alter signal paths. Different human activities produce unique CSI signatures, enabling accurate activity recognition using advanced signal processing and machine learning methods. 

Skills and tools: MATLAB, Python, scikit-learn, pandas, machine learning, signal processing, time series, contrastive learning, knowledge distillation, SimCLR, MiniROCKET, Raspberry Pi

Publications:

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

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

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

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

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

Hypertension, commonly known as high blood pressure, poses a serious health risk as it often manifests without noticeable symptoms, silently damaging internal organs. Recognizing the need for a user-friendly system that provides precise and continuous blood pressure monitoring without the inconvenience of cuff-based methods, this project aimed to develop a blood pressure estimation Android application. This application estimates both systolic and diastolic blood pressures by analyzing Electrocardiogram (ECG) and Photoplethysmogram (PPG) signals received via USB or Bluetooth. It extracts relevant features such as Pulse Arrival Time (PAT) and utilizes pre-trained machine learning models including Random Forest and AdaBoost for accurate predictions.

Skills and tools: MATLAB, Python, Java, Android Studio, scikit-learn, pandas, HSMMlearn, machine learning, signal processing, Arduino

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

Based on a World Health Organization (WHO) report from July 2020, the COVID-19 pandemic, caused by SARS-CoV-2, primarily affects the respiratory system. The spread and impact of this infection can be modeled using mathematical differential equations. Furthermore, deep neural networks have made significant advances, achieving high accuracy in tasks such as computer vision and medical diagnosis. Our research demonstrates that convolutional deep neural networks are promising tools for diagnosing COVID-19 using CT scans by localizing and segmenting lesions, thereby reducing diagnostic errors and aiding physicians, particularly in the early stages of infection. Additionally, we found that deep learning techniques applied to chest X-rays can accurately classify COVID-19, distinguishing it from other types of pneumonia and healthy lungs. These AI models improve the speed and accuracy of diagnoses, facilitating rapid COVID-19 identification and enhancing patient care.

Skills and tools: Python, scikit-learn, pandas, machine learning, signal processing, deep learning, PyTorch, Keras, Tensorflow, U-Net, VGG, EfficientNet, and ResNet

Publications:

1- A fractional-order model for the novel coronavirus (COVID-19) outbreak

2- Segmentation of COVID-19 Infections on CT: Comparison of Four UNet-Based Networks (dataset)

3- Detection of COVID-19 from Chest Radiographs: Comparison of Four End-to-End Trained Deep Learning Models

4- A Multi-centric Evaluation of Deep Learning Models for Segmentation of COVID-19 Lung Lesions on Chest CT Scans (code)

5- Accuracy of artificial intelligence CT quantification in predicting COVID-19 subjects’ prognosis

Implementing fast depth-wise Convolutional Neural Networks (CNNs) for detecting car plates, models, and speeds on Raspberry Pi devices involved leveraging the efficiency of depth-wise separable convolutions. These networks, optimized for the limited computational power of Raspberry Pi, significantly enhanced real-time image processing capabilities. By employing C++ for implementation, we successfully maximized the system's performance and responsiveness, ensuring rapid detection and analysis. This approach enabled robust, low-latency applications in smart traffic monitoring and automated toll collection, providing an efficient and scalable solution for edge computing environments. Additionally, using Raspberry Pi devices offers a cost-efficient alternative to more expensive systems, making it accessible for widespread deployment. This method is now currently used in various cities for traffic monitoring and real-time car speed measurement, demonstrating both its effectiveness and economic advantage.

Skills and tools: C++, Python, PyTorch, OpenCV, Qt, machine learning, deep learning, Raspberry Pi, Depth-wise CNN