A Wearable Multimodal Sensing System for Classification of Mental and Physical Stress States
Disciplines
Biomedical
Abstract (300 words maximum)
Stress is a physiological and psychological state that occurs when the body responds to internal or external stimuli. It can be broadly categorized into mental stress, which arises from cognitive pressure, and physical stress, which stems from bodily exertion or environmental factors. Both forms of stress affect multiple biological systems and can influence long-term health outcomes if not properly managed. This raises the central research question of how mental and physical stress, given their different effects, can be reliably classified to support improved health monitoring, early intervention, and personalized well-being optimization. To address this, a wrist-worn multimodal sensing system is presented that integrates non-invasive sensors such as photoplethysmography (PPG), accelerometer, skin temperature sensor, electromyography (EMG), and electrodermal activity (EDA) within a sensor fusion framework. The fusion algorithm processes these heterogeneous signals through pre-processing, feature extraction, and pattern recognition to identify stress-related physiological changes. It combines synchronized features from both time and frequency domains, providing a more comprehensive representation of stress responses than single-mode sensing. The system operates through continuous monitoring and feature-level fusion, followed by classification into four discrete states, including physical stress, mental stress, combined stress, and no-stress conditions. The proposed model enhances discrimination accuracy and adaptability across varying conditions. Experimental evaluations conducted under controlled physical and mental stress conditions demonstrate its effectiveness in differentiating two stress types and intensity levels. Unlike existing research and commercial-grade devices that employ multimodal sensing mainly for general stress detection by excluding some daily activities to simplify system and maintain generalization, this system introduces a classification-centered framework. The novel approach is optimized for faster and continuous operation, supporting real-time stress assessment in dynamic and everyday environments. This advancement represents a step toward wearable systems, capable of delivering personalized, context-aware, and continuous stress monitoring for proactive health and wellness applications.
Use of AI Disclaimer
no
Academic department under which the project should be listed
SPCEET – Electrical and Computer Engineering
Primary Investigator (PI) Name
Razvan Cristian Voicu
Presentation Slide
A Wearable Multimodal Sensing System for Classification of Mental and Physical Stress States
Stress is a physiological and psychological state that occurs when the body responds to internal or external stimuli. It can be broadly categorized into mental stress, which arises from cognitive pressure, and physical stress, which stems from bodily exertion or environmental factors. Both forms of stress affect multiple biological systems and can influence long-term health outcomes if not properly managed. This raises the central research question of how mental and physical stress, given their different effects, can be reliably classified to support improved health monitoring, early intervention, and personalized well-being optimization. To address this, a wrist-worn multimodal sensing system is presented that integrates non-invasive sensors such as photoplethysmography (PPG), accelerometer, skin temperature sensor, electromyography (EMG), and electrodermal activity (EDA) within a sensor fusion framework. The fusion algorithm processes these heterogeneous signals through pre-processing, feature extraction, and pattern recognition to identify stress-related physiological changes. It combines synchronized features from both time and frequency domains, providing a more comprehensive representation of stress responses than single-mode sensing. The system operates through continuous monitoring and feature-level fusion, followed by classification into four discrete states, including physical stress, mental stress, combined stress, and no-stress conditions. The proposed model enhances discrimination accuracy and adaptability across varying conditions. Experimental evaluations conducted under controlled physical and mental stress conditions demonstrate its effectiveness in differentiating two stress types and intensity levels. Unlike existing research and commercial-grade devices that employ multimodal sensing mainly for general stress detection by excluding some daily activities to simplify system and maintain generalization, this system introduces a classification-centered framework. The novel approach is optimized for faster and continuous operation, supporting real-time stress assessment in dynamic and everyday environments. This advancement represents a step toward wearable systems, capable of delivering personalized, context-aware, and continuous stress monitoring for proactive health and wellness applications.