Deteksi Stres Non-Kontak Menggunakan Integrasi Radar Continuous Wave dan Kamera Berbasis Machine Learning

Lestiana, Siska Dwi (2026) Deteksi Stres Non-Kontak Menggunakan Integrasi Radar Continuous Wave dan Kamera Berbasis Machine Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221017-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Stres merupakan masalah kesehatan yang banyak dialami masyarakat dan apabila tidak dikenali sejak dini dapat berkembang menjadi kondisi yang berdampak pada kesehatan fisik, fungsi kognitif, dan kesejahteraan psikologis. Oleh karena itu, diperlukan metode deteksi stres yang praktis dan nyaman untuk mendukung pemantauan secara berkelanjutan. Penelitian ini bertujuan mengembangkan sistem deteksi stres non-kontak berbasis radar Continuous Wave (CW) 24 GHz dan kamera Raspberry Pi dengan mengintegrasikan informasi fisiologis kardiorespirasi dan ekspresi wajah. Sinyal radar diproses menggunakan Extended Differentiate and Cross-Multiply (Extended-DACM), kemudian dipisahkan menjadi komponen jantung dan pernapasan menggunakan Health-Variational Mode Decomposition (Health-VMD). Dari kedua komponen tersebut diekstraksi fitur Heart Rate Variability (HRV) dan Respiratory Rate Variability (RRV), sedangkan ekspresi wajah dianalisis menggunakan OpenFace 2.2.0 untuk memperoleh fitur Action Units (AU). Kanal radar menggunakan ensemble classifier yang terdiri atas Logistic Regression, Linear Discriminant Analysis (LDA) dengan regularisasi shrinkage, dan Gaussian Naïve Bayes, sedangkan kanal wajah menggunakan Gaussian Naïve Bayes. Probabilitas kedua kanal kemudian digabungkan melalui pendekatan late fusion berbasis rata-rata logit terbobot (weighted logit averaging), dengan bobot dioptimasi melalui nested cross-validation. Evaluasi menggunakan skema Leave-One-Subject-Out terhadap 17 subjek menunjukkan bahwa model radar memperoleh akurasi 68,2% (ROC-AUC 0,715), model wajah memperoleh akurasi 50,0% (ROC-AUC 0,462), sedangkan model late fusion memberikan performa terbaik dengan akurasi 71,2%, F1-score 74,0%, dan ROC-AUC 0,713. Hasil tersebut menunjukkan bahwa integrasi informasi fisiologis dan visual mampu meningkatkan kemampuan sistem dalam membedakan kondisi stres dan relaksasi secara non-kontak.
=========================================================================================================================================
Stress is a common health issue that affects many people and, if not identified at an early stage, may develop into a condition that adversely impacts physical health, cognitive function, and psychological well-being. Therefore, a practical and comfortable stress detection method is needed to support continuous monitoring. This study aims to develop a non-contact stress detection system based on a 24 GHz Continuous Wave (CW) radar and a Raspberry Pi camera by integrating cardiorespiratory physiological information and facial expressions. Radar signals were processed using the Extended Differentiate and Cross-Multiply (Extended-DACM) algorithm and subsequently separated into cardiac and respiratory components using Health-Variational Mode Decomposition (Health-VMD). Heart Rate Variability (HRV) and Respiratory Rate Variability (RRV) features were extracted from these components, while facial expressions were analyzed using OpenFace 2.2.0 to obtain Action Unit (AU) features. The radar branch employed an ensemble classifier consisting of Logistic Regression, shrinkage-regularized Linear Discriminant Analysis (LDA), and Gaussian Naïve Bayes, whereas the facial branch used Gaussian Naïve Bayes. The output probabilities from both branches were then combined using a late fusion approach based on weighted logit averaging, with the weights optimized through nested cross-validation. Evaluation using the Leave-One-Subject-Out (LOSO) scheme on 17 subjects showed that the radar model achieved an accuracy of 68.2% (ROC-AUC = 0.715), the facial model achieved an accuracy of 50.0% (ROC-AUC = 0.462), while the late fusion model achieved the best performance with an accuracy of 71.2%, an F1-score of 74.0%, and a ROC-AUC of 0.713. These results demonstrate that integrating physiological and visual information improves the system's capability to distinguish between stress and relaxation conditions in a non-contact manner.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi stres, Radar Continuous Wave, Action Units, Heart Rate Variability, Respiratory Rate Variability, Late Fusion; Stress detection, Continuous Wave radar, Action Units, Heart Rate Variability, Respiratory Rate Variability, Late Fusion.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Industrial Technology > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Siska Dwi Lestiana
Date Deposited:	31 Jul 2026 08:06
Last Modified:	31 Jul 2026 08:06
URI:	http://repository.its.ac.id/id/eprint/141039

Actions (login required)

View Item