PPG-Posture: Wearable Posture Recognition from Photoplethysmography Signals
Project Description
Photoplethysmography, or PPG, is a non-invasive optical sensing technique widely available in smartwatches and other wearable devices. It is commonly used to measure physiological indicators such as heart rate, heart rate variability, and blood oxygen saturation. Beyond these conventional applications, changes in body posture may also influence the morphology and characteristics of the PPG waveform.
Recognizing human body posture can support applications such as sleep monitoring, elderly care, workplace ergonomics, and personal health management. Existing posture-recognition systems often rely on cameras, multiple inertial sensors, or pressure-sensing infrastructure, which may introduce concerns related to privacy, device placement, or deployment complexity. This project investigates whether body posture can be inferred primarily from PPG signals already collected by common wearable devices.
The project will explore the collection and analysis of PPG signals under common sitting, lying, and daily-life postures. Possible research components include signal preprocessing, waveform-quality assessment, morphology-based feature extraction, feature or representation learning, and machine-learning-based posture recognition. The project may also investigate robustness across users, recording sessions, wearable placements, and activity conditions. Depending on progress, the work may be extended toward posture-transition detection, continuous monitoring, or lightweight real-time implementation on wearable devices.
The long-term objective is to explore how an existing physiological sensor can be used not only for monitoring vital signs but also for understanding the user’s physical context, enabling simpler and more integrated wearable health-sensing systems.
Recognizing human body posture can support applications such as sleep monitoring, elderly care, workplace ergonomics, and personal health management. Existing posture-recognition systems often rely on cameras, multiple inertial sensors, or pressure-sensing infrastructure, which may introduce concerns related to privacy, device placement, or deployment complexity. This project investigates whether body posture can be inferred primarily from PPG signals already collected by common wearable devices.
The project will explore the collection and analysis of PPG signals under common sitting, lying, and daily-life postures. Possible research components include signal preprocessing, waveform-quality assessment, morphology-based feature extraction, feature or representation learning, and machine-learning-based posture recognition. The project may also investigate robustness across users, recording sessions, wearable placements, and activity conditions. Depending on progress, the work may be extended toward posture-transition detection, continuous monitoring, or lightweight real-time implementation on wearable devices.
The long-term objective is to explore how an existing physiological sensor can be used not only for monitoring vital signs but also for understanding the user’s physical context, enabling simpler and more integrated wearable health-sensing systems.
Supervisor
SHAO, Qijia
Quota
2
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP3200
UROP4100
Applicant's Roles
The applicant will participate in different stages of the research project. Possible responsibilities include:
Reviewing related work in PPG sensing, wearable computing, signal processing, and posture recognition.
Supporting the design of data-collection procedures for different body postures and daily activities.
Assisting in the development or configuration of a wearable PPG data-collection system.
Conducting pilot experiments and supporting data collection, annotation, organization, and quality checking.
Implementing signal-processing methods for filtering, waveform segmentation, morphology analysis, and feature extraction.
Developing and evaluating machine-learning models for posture recognition.
Investigating model robustness across participants, sessions, and sensing conditions.
Contributing to prototype development, result interpretation, research documentation, and potential paper preparation.
The applicant’s exact responsibilities may be adjusted according to their background, interests, and the progress of the project.
Reviewing related work in PPG sensing, wearable computing, signal processing, and posture recognition.
Supporting the design of data-collection procedures for different body postures and daily activities.
Assisting in the development or configuration of a wearable PPG data-collection system.
Conducting pilot experiments and supporting data collection, annotation, organization, and quality checking.
Implementing signal-processing methods for filtering, waveform segmentation, morphology analysis, and feature extraction.
Developing and evaluating machine-learning models for posture recognition.
Investigating model robustness across participants, sessions, and sensing conditions.
Contributing to prototype development, result interpretation, research documentation, and potential paper preparation.
The applicant’s exact responsibilities may be adjusted according to their background, interests, and the progress of the project.
Applicant's Learning Objectives
Develop an understanding of PPG sensing and the physiological and physical factors that affect PPG waveform morphology.
Gain hands-on experience in wearable sensing system development and human-participant data collection.
Learn fundamental techniques for physiological signal preprocessing, waveform analysis, and feature extraction.
Apply machine-learning methods to real-world wearable sensor data.
Understand how to design and evaluate models across different users, sessions, and environmental conditions.
Gain experience in experimental design, data-quality assessment, and responsible handling of human-participant data.
Improve research communication, teamwork, technical documentation, and scientific presentation skills.
Gain hands-on experience in wearable sensing system development and human-participant data collection.
Learn fundamental techniques for physiological signal preprocessing, waveform analysis, and feature extraction.
Apply machine-learning methods to real-world wearable sensor data.
Understand how to design and evaluate models across different users, sessions, and environmental conditions.
Gain experience in experimental design, data-quality assessment, and responsible handling of human-participant data.
Improve research communication, teamwork, technical documentation, and scientific presentation skills.
Complexity of the project
Moderate