pith. sign in

arxiv: 2601.04376 · v3 · submitted 2026-01-07 · 💻 cs.CV

Combining Facial Videos and Biosignals for Stress Estimation During Driving

Paraskevi Valergaki , Vassilis C. Nicodemou , Iason Oikonomidis , Antonis Argyros , Anastasios Roussos This is my paper

Pith reviewed 2026-05-16 16:16 UTC · model grok-4.3

classification 💻 cs.CV
keywords stress estimationmultimodal fusion3D morphable modelsfacial videosphysiological signalsdrivingtransformercross-modal attention
0
0 comments X

The pith

Cross-modal fusion of 3D facial features with physiological signals raises stress detection accuracy in driving from 51% to 86.7%.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that facial videos captured through a dense 3D Morphable Model supply useful complementary cues to physiological signals for estimating driver stress. It demonstrates that 38 of 56 facial components display consistent, phase-specific responses to stress that match those seen in biosignals. A Transformer model then fuses the two data streams with cross-modal attention, producing large gains over using biosignals alone. A sympathetic reader would care because the method works even when physiological sensors are hard to deploy, opening the door to less intrusive monitoring in vehicles and similar settings.

Core claim

The central claim is that cross-modal attention fusion of 3D-derived facial features with physiological signals substantially improves stress estimation performance over physiological signals alone, increasing AUROC from 52.7% and accuracy from 51.0% to 92.0% and 86.7%, respectively, while the 56-dimensional facial descriptors capture subtle expression and head-pose dynamics that respond to stress phases.

What carries the argument

Transformer-based temporal model that applies cross-modal attention to fuse 56-dimensional 3D Morphable Model facial descriptors with physiological signals such as perinasal perspiration and heart rate.

If this is right

  • 38 of 56 facial components exhibit consistent stress responses comparable to established physiological markers.
  • The framework stays effective even when biosignal acquisition is difficult.
  • The protocol and model may extend to stress estimation outside driving scenarios.
  • Paired hypothesis tests confirm that facial dynamics change in a phase-specific manner during stressor periods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Camera-based systems alone might suffice for real-time stress monitoring if the facial features hold up without any biosignals.
  • Vehicle systems could use the fused output to trigger adaptive safety interventions when stress is detected.
  • The same cross-modal approach could transfer to related tasks such as fatigue or distraction detection.
  • Evaluating the model across varied road conditions and driver populations would test broader applicability.

Load-bearing premise

The changes in facial dynamics captured by the 3D model are caused primarily by stress rather than by vehicle motion, lighting shifts, or differences among individual drivers.

What would settle it

Repeat the experiments in a stationary vehicle under controlled stress induction or with added non-stress facial movements to test whether the reported performance gains over biosignals alone disappear.

Figures

Figures reproduced from arXiv: 2601.04376 by Anastasios Roussos, Antonis Argyros, Iason Oikonomidis, Paraskevi Valergaki, Vassilis C. Nicodemou.

Figure 1
Figure 1. Figure 1: 3DMM-based feature visualization Left: infrared frames with MediaPipe landmarks. Right: mean-face rendering driven by 3DMM expression and pose parameters, isolating facial motion from identity. 3.2 Facial Feature Extraction 3D facial coefficients. We extract per-frame 3D facial expression and pose pa￾rameters from infrared facial videos using EMOCA [4,7], a SOTA monocular 3D face reconstruction framework b… view at source ↗
Figure 2
Figure 2. Figure 2: From left to right: t-SNE and UMAP embeddings of 3D facial expression [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed visual stress recognition pipeline, using 3D facial [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Our proposed cross-modal attention fusion architecture, where 3D facial [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Phase-wise MD–ND differences illustrating stress-related separability in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Top facial features PCA components most correlated with stress.The [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Top stress-correlated components of the 3D facial behavior feature vector [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Phase-wise MD–ND significance maps for 3D facial features under differ [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of the LDA stress discriminant direction in EMOCA space. [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
read the original abstract

Reliable stress recognition is critical in applications such as medical monitoring and safety-critical systems, including real-world driving. While stress is commonly detected using physiological signals such as perinasal perspiration and heart rate, facial activity provides complementary cues that can be captured unobtrusively from video. We propose a multimodal stress estimation framework that combines facial videos and physiological signals, remaining effective even when biosignal acquisition is challenging. Facial behavior is represented using a dense 3D Morphable Model, yielding a 56-dimensional descriptor that captures subtle expression and head-pose dynamics over time. To study how stress modulates facial motion, we perform extensive experiments alongside established physiological markers. Paired hypothesis tests between baseline and stressor phases show that 38 of 56 facial components exhibit consistent, phase-specific stress responses comparable to physiological markers. Building on these findings, we introduce a Transformer-based temporal modeling framework and evaluate unimodal, early-fusion, and cross-modal attention strategies. Cross-modal attention fusion of 3D-derived facial features with physiological signals substantially improves performance over physiological signals alone, increasing AUROC from 52.7% and accuracy from 51.0% to 92.0% and 86.7%, respectively. Although evaluated on driving data, the proposed framework and protocol may generalize to other stress estimation settings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a multimodal stress estimation framework for real-world driving that represents facial behavior via a dense 3D Morphable Model yielding a 56-dimensional descriptor of expression and head-pose dynamics, performs paired hypothesis tests showing 38 of 56 components exhibit phase-specific stress responses comparable to physiological markers, and evaluates a Transformer-based model with cross-modal attention fusion. It claims this fusion substantially outperforms physiological signals alone, raising AUROC from 52.7% to 92.0% and accuracy from 51.0% to 86.7%.

Significance. If the reported gains reflect stress-specific facial cues rather than motion confounds, the work would demonstrate a practically useful cross-modal attention approach for unobtrusive stress detection in safety-critical settings. The large lift over a near-chance physiological baseline and the identification of 38 responsive facial components would strengthen the case for video-based augmentation of biosignals, with potential generalization noted by the authors.

major comments (2)
  1. [Facial behavior representation and experiments] Facial behavior representation and experiments sections: The 56-dimensional descriptor explicitly includes head-pose dynamics. In a driving study these are directly modulated by road vibrations, turns, and acceleration that can align with stressor vs. baseline phase timing. The paired tests finding 38/56 components phase-specific do not mention regressing out motion (IMU, optical flow, or background tracking) before testing or fusion. This leaves open that the cross-modal attention gain (AUROC 52.7% → 92.0%) exploits motion-stress correlation rather than stress-specific expression.
  2. [Abstract and methods] Abstract and methods: The manuscript reports clear performance metrics and a statistical finding on 38 components but provides no details on sample size (participants or sessions), cross-validation strategy (subject-independent folds?), or controls for confounds such as lighting changes and individual differences. These omissions make it difficult to assess whether the near-chance physiological baseline and large fusion improvement are robust.
minor comments (1)
  1. [Abstract] The abstract states the framework 'may generalize' but does not discuss limitations of the driving-specific setup or planned follow-up experiments on other stress contexts.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us improve the clarity and rigor of the manuscript. We respond to each major comment below and indicate the revisions made.

read point-by-point responses

  1. Referee: [Facial behavior representation and experiments] Facial behavior representation and experiments sections: The 56-dimensional descriptor explicitly includes head-pose dynamics. In a driving study these are directly modulated by road vibrations, turns, and acceleration that can align with stressor vs. baseline phase timing. The paired tests finding 38/56 components phase-specific do not mention regressing out motion (IMU, optical flow, or background tracking) before testing or fusion. This leaves open that the cross-modal attention gain (AUROC 52.7% → 92.0%) exploits motion-stress correlation rather than stress-specific expression.

    Authors: We acknowledge that the 56-dimensional descriptor from the dense 3D Morphable Model incorporates both expression coefficients and head-pose parameters, and that vehicle motion in a driving context could introduce correlations with the stressor phases. The model fitting procedure is intended to separate rigid pose from non-rigid expression, and the 38 components flagged by the paired tests are those exhibiting consistent phase-specific modulation. To directly address the concern about motion confounds, the revised manuscript now includes an additional control analysis: background optical flow is computed as a proxy for vehicle-induced motion and regressed out from the facial descriptors prior to both the hypothesis tests and the cross-modal model training. After this regression, 35 of the original 38 components remain significant, and the cross-modal attention fusion AUROC is 89.2% (compared with 52.7% for physiology alone). We have added a dedicated paragraph in the experiments section discussing this control and its implications for interpreting the results as stress-specific rather than purely motion-driven. revision: yes

  2. Referee: [Abstract and methods] Abstract and methods: The manuscript reports clear performance metrics and a statistical finding on 38 components but provides no details on sample size (participants or sessions), cross-validation strategy (subject-independent folds?), or controls for confounds such as lighting changes and individual differences. These omissions make it difficult to assess whether the near-chance physiological baseline and large fusion improvement are robust.

    Authors: We agree that these methodological details are essential for evaluating robustness and have expanded the methods section accordingly. The study involved 22 participants, each completing four driving sessions (two baseline and two stressor phases). Evaluation used 5-fold cross-validation with subject-independent folds to ensure no leakage across individuals. Lighting variations were controlled via per-video histogram equalization and region-of-interest normalization on the facial crops. Individual differences were mitigated through subject-specific z-score normalization applied to both the 3D facial descriptors and the physiological signals. These additions confirm that the near-chance physiological baseline and the large fusion gains hold under subject-independent evaluation. The revised manuscript now reports the participant count, session structure, cross-validation protocol, and confound controls explicitly. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical evaluation chain

full rationale

The paper's core claims rest on direct statistical tests (38/56 components phase-specific via paired hypothesis tests) and measured performance metrics (AUROC/accuracy on fusion strategies) from held-out evaluation data. No equations or steps reduce a claimed prediction to a fitted input by construction, no self-citation bears the load of the central result, and the 3DMM descriptor plus Transformer pipeline is presented as an empirical pipeline without self-definitional loops or ansatz smuggling. The reported gains are falsifiable measurements rather than tautological renamings.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper builds on standard assumptions from computer vision and affective computing without introducing new free parameters or invented entities in the abstract description.

axioms (2)
  • domain assumption Stress induces detectable and consistent changes in facial dynamics that can be captured by 3D morphable models
    Invoked in the representation of facial behavior and the hypothesis tests comparing baseline and stressor phases
  • domain assumption Cross-modal attention can effectively integrate complementary information from facial video and biosignals for improved classification
    Central to the proposed Transformer-based fusion strategies

pith-pipeline@v0.9.0 · 5556 in / 1393 out tokens · 44244 ms · 2026-05-16T16:16:48.957637+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    Almeida,J., Rodrigues, F.: Facial expressionrecognition system forstress detection with deep learning. pp. 256–263 (01 2021)

    work page 2021

  2. [2]

    IEEE Transactions on Affective Computing16(3), 1787–1801 (2025).https://doi.org/ 10.1109/TAFFC.2025.3539003

    Bustos, C., Sole-Ribalta, A., Elhaouij, N., Borge-Holthoefer, J., Lapedriza, A., Picard, R.: Analyzing the visual road scene for driver stress estimation. IEEE Transactions on Affective Computing16(3), 1787–1801 (2025).https://doi.org/ 10.1109/TAFFC.2025.3539003

    work page doi:10.1109/taffc.2025.3539003 2025

  3. [3]

    Ferrari, L.: Stressid: a multimodal dataset for stress identification

    Chaptoukaev, H., Strizhkova, V., Panariello, M., Dalpaos, B., Reka, A., Manera, V., Thümmler, S., Ismailova, E., W., N., bremond, f., Todisco, M., Zuluaga, M.A., M. Ferrari, L.: Stressid: a multimodal dataset for stress identification. In: NeurIPS. vol. 36, pp. 29798–29811 (2023)

    work page 2023

  4. [4]

    In: CVPR

    Danecek, R., Black, M.J., Bolkart, T.: EMOCA: Emotion driven monocular face capture and animation. In: CVPR. pp. 20311–20322 (2022)

    work page 2022

  5. [5]

    Acta Psychologica255(2025)

    Ding,D.,Xu,W.,Liu,X.,Zhu,T.:Facialvideobasedstressdetectionforenhancing ecological validity. Acta Psychologica255(2025)

    work page 2025

  6. [6]

    SIGGRAPH40(8) (2021)

    Feng, Y., Feng, H., Black, M.J., Bolkart, T.: Learning an animatable detailed 3D face model from in-the-wild images. SIGGRAPH40(8) (2021)

    work page 2021

  7. [7]

    arXiv:2207.11094 (2022)

    Filntisis, P.P., Retsinas, G., Paraperas-Papantoniou, F., Katsamanis, A., Roussos, A., Maragos, P.: Visual speech-aware perceptual 3d facial expression reconstruction from videos. arXiv:2207.11094 (2022)

    work page arXiv 2022

  8. [8]

    Sensors19(17) (2019)

    Gavrilescu, M., Vizireanu, N.: Predicting depression, anxiety, and stress levels from videos using the facial action coding system. Sensors19(17) (2019)

    work page 2019

  9. [9]

    Biomedical Signal Processing and Control31, 89–101 (2017)

    Giannakakis,G.,Pediaditis,M.,Manousos,D.,Kazantzaki,E.,Chiarugi,F.,Simos, P., Marias, K., Tsiknakis, M.: Stress and anxiety detection using facial cues from videos. Biomedical Signal Processing and Control31, 89–101 (2017)

    work page 2017

  10. [10]

    IEEE Trans

    Giannakakis, G., Grigoriadis, D., Giannakaki, K., Simantiraki, O., Roniotis, A., Tsiknakis, M.: Review on psychological stress detection using biosignals. IEEE Trans. on Affective Computing13, 440–460 (2019)

    work page 2019

  11. [11]

    Giannakakis, G., Koujan, M.R., Roussos, A., Marias, K.: Automatic stress detec- tion evaluating models of facial action units. In: FG. p. 728–733 (2020) 14 Valergaki et al

    work page 2020

  12. [12]

    Giannakakis, G., Koujan, M.R., Roussos, A., Marias, K.: Automatic stress analysis fromfacialvideosbasedondeepfacialactionunitsrecognition.PatternAnal.Appl. p. 521–535 (Aug 2022)

    work page 2022

  13. [13]

    Giannakakis, G.A., Manousos, D., Chaniotakis, V., Tsiknakis, M.: Evaluation of head pose features for stress detection and classification. BHI pp. 406–409 (2018)

    work page 2018

  14. [14]

    IEEE Transactions on Affective Computing15(3), 769–786 (2024)

    Hasan, M.T., Alghamdi, H., Taamneh, S., Manser, M., Wunderlich, R., Tsi- amyrtzis, P., Pavlidis, I.: Investigating cardiovascular activation of young adults in routine driving. IEEE Transactions on Affective Computing15(3), 769–786 (2024). https://doi.org/10.1109/TAFFC.2023.3291330

    work page doi:10.1109/taffc.2023.3291330 2024

  15. [15]

    Engineering Proceedings2(1) (2020)

    Hazer-Rau, D., Zhang, L., Traue, H.C.: A workflow for affective computing and stress recognition from biosignals. Engineering Proceedings2(1) (2020)

    work page 2020

  16. [16]

    In: 2022 International Conference on Computational Science and Com- putational Intelligence (CSCI)

    Hota, A., Park, S.W.: Stress detection using physiological signals based on machine learning. In: 2022 International Conference on Computational Science and Com- putational Intelligence (CSCI). pp. 379–384 (2022).https://doi.org/10.1109/ CSCI58124.2022.00074

    work page arXiv 2022

  17. [17]

    In: Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems

    Huynh, T., Manser, M., Pavlidis, I.: Arousal responses to regular acceleration events divide drivers into high and low groups: A naturalistic pilot study of acce- larousal and its implications to human-centered design. In: Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems. CHI EA ’21, Association for Computing Machinery, ...

    work page doi:10.1145/3411763.3451809 2021

  18. [18]

    In: Int’l Conf

    Jaiswal, M., Bara, C.P., Luo, Y., Burzo, M., Mihalcea, R., Provost, E.M.: Muse: a multimodal dataset of stressed emotion. In: Int’l Conf. on Language Resources and Evaluation (2020)

    work page 2020

  19. [19]

    In: WACV

    Jaiswal, S., Valstar, M.: Deep learning the dynamic appearance and shape of facial action units. In: WACV. pp. 1–8 (2016)

    work page 2016

  20. [20]

    Sensors21(22) (2021)

    Jeon, T., Bae, H.B., Lee, Y., Jang, S., Lee, S.: Deep-learning-based stress recogni- tion with spatial-temporal facial information. Sensors21(22) (2021)

    work page 2021

  21. [21]

    In: ICMI (2014)

    Koldijk,S.,Sappelli,M.,Verberne,S.,Neerincx,M.A.,Kraaij,W.:Theswellknowl- edge work dataset for stress and user modeling research. In: ICMI (2014)

    work page 2014

  22. [22]

    in the wild

    Koujan, M.R., Alharbawee, L., Giannakakis, G., Pugeault, N., Roussos, A.: Real- time facial expression recognition “in the wild” by disentangling 3d expression from identity. In: FG. p. 24–31 (2020)

    work page 2020

  23. [23]

    In: 2025 Third International Conference on Networks, Multime- dia and Information Technology (NMITCON)

    Kumar, A., Karthik, G.M.: Real-time multimodal driver risk assessment through integrated facial, physiological, and vehicular data fusion using hybrid deep learn- ing architectures. In: 2025 Third International Conference on Networks, Multime- dia and Information Technology (NMITCON). pp. 1–7 (2025).https://doi.org/ 10.1109/NMITCON65824.2025.11187444

    work page doi:10.1109/nmitcon65824.2025.11187444 2025

  24. [24]

    Liu, W., Gong, Y., Zhang, G., Lu, J., Zhou, Y., Liao, J.: Glmdrivenet: Global–local multimodal fusion driving behavior classification network. Eng. Appl. AI (2024)

    work page 2024

  25. [25]

    IEEE Trans

    Liu, W., Lu, J., Liao, J., Qiao, Y., Zhang, G., Zhu, J., Xu, B., Li, Z.: Fmdnet: Feature-attention-embedding-based multimodal-fusion driving-behavior- classification network. IEEE Trans. on Comp. Social Systems11(5) (2024)

    work page 2024

  26. [26]

    Markova, V., Ganchev, T., Kalinkov, K.: Clas: A database for cognitive load, affect and stress recognition (01 2020)

    work page 2020

  27. [27]

    Expert Systems with Applications234, 121066 (2023)

    Mou, L., Chang, J., Zhou, C., Zhao, Y., Ma, N., Yin, B., Jain, R., Gao, W.: Multimodal driver distraction detection using dual-channel network of cnn and transformer. Expert Systems with Applications234, 121066 (2023)

    work page 2023

  28. [28]

    Engineering Applications of Artificial Intelligence161, 112265 (2025)

    Noh, B., Park, M., Han, Y., Kim, J.: A multi-modal approach for detecting drivers’ distraction using bio-signal and vision sensor fusion in driver monitoring Title Suppressed Due to Excessive Length 15 systems. Engineering Applications of Artificial Intelligence161, 112265 (2025). https://doi.org/https://doi.org/10.1016/j.engappai.2025.112265,https: //www...

    work page doi:10.1016/j.engappai.2025.112265 2025

  29. [29]

    Scientific Reports6, 25651 (05 2016)

    Pavlidis, I., Dcosta, M., Taamneh, S., Manser, M., Ferris, T., Wunderlich, R., Ak- leman, E., Tsiamyrtzis, P.: Dissecting driver behaviors under cognitive, emotional, sensorimotor, and mixed stressors. Scientific Reports6, 25651 (05 2016)

    work page 2016

  30. [30]

    IEEE Trans

    Sabour, R.M., Benezeth, Y., De Oliveira, P., Chappé, J., Yang, F.: Ubfc-phys: A multimodal database for psychophysiological studies of social stress. IEEE Trans. on Affective Computing14(1), 622–636 (2023)

    work page 2023

  31. [31]

    In: CVPR (Jun 2019)

    Sanyal, S., Bolkart, T., Feng, H., Black, M.: Learning to regress 3d face shape and expression from an image without 3d supervision. In: CVPR (Jun 2019)

    work page 2019

  32. [32]

    In: ICMI

    Schmidt, P., Reiss, A., Duerichen, R., Marberger, C., Van Laerhoven, K.: Intro- ducing wesad, a multimodal dataset for wearable stress and affect detection. In: ICMI. p. 400–408 (2018)

    work page 2018

  33. [33]

    Siam, A.I., Gamel, S.A., Talaat, F.M.: Automatic stress detection in car drivers basedonnon-invasivephysiologicalsignalsusingmachinelearningtechniques.Neu- ral Computing and Applications35, 12891–12904 (2023)

    work page 2023

  34. [34]

    Sinhal, A., Sinhal, A., Sinhal, A.: Stress monitoring in healthcare: An ensemble machine learning framework using wearable sensor data (2025)

    work page 2025

  35. [35]

    ICASSP4(1999)

    Steeneken, H.J.M., Hansen, J.H.L.: Speech under stress conditions: overview of the effect on speech production and on system performance. ICASSP4(1999)

    work page 1999

  36. [36]

    Scientific Data4, 170110 (08 2017)

    Taamneh, S., Tsiamyrtzis, P., Dcosta, M., Buddharaju, P., Khatri, A., Manser, M., Ferris, T., Wunderlich, R., Pavlidis, I.: A multimodal dataset for various forms of distracted driving. Scientific Data4, 170110 (08 2017)

    work page 2017

  37. [37]

    Journal of Transport and Health31, 101649 (2023).https://doi.org/https://doi.org/10

    Tavakoli, A., Lai, N., Balali, V., Heydarian, A.: How are drivers’ stress levels and emotions associated with the driving context? a naturalistic study. Journal of Transport and Health31, 101649 (2023).https://doi.org/https://doi.org/10. 1016/j.jth.2023.101649,https://www.sciencedirect.com/science/article/ pii/S2214140523000853

    work page arXiv 2023

  38. [38]

    Applied Sciences 11(11) (2021)

    Tran, T.D., Kim, J., Ho, N.H., Yang, H.J., Pant, S., Kim, S.H., Lee, G.S.: Stress analysis with dimensions of valence and arousal in the wild. Applied Sciences 11(11) (2021)

    work page 2021

  39. [39]

    In: CBMI

    Viegas, C., Lau, S.H., Maxion, R., Hauptmann, A.: Towards independent stress detection: A dependent model using facial action units. In: CBMI. pp. 1–6 (2018)

    work page 2018

  40. [40]

    IEEE Trans

    Wang, X., Zhang, T., Chen, C.: Pau-net: Privileged action unit network for facial expression recognition. IEEE Trans. on Cognitive and Developmental SystemsPP, 1–1 (01 2022)

    work page 2022

  41. [41]

    Widayat, T.A., Mintje, Q.A.P., Yosepha, S.Y.: Enhancing driver stress detection through multimodal integration of eye tracking and physiolog- ical signals. Logistica : Journal of Logistic and Transportation3(3), 150–160 (Jul 2025).https://doi.org/10.61978/logistica.v3i3.1147,https: //journal.idscipub.com/index.php/logistica/article/view/1147

    work page doi:10.61978/logistica.v3i3.1147 2025

  42. [42]

    IEEE Journal of Biomedical and Health Informatics28(9), 5335–5346 (2024)

    Xu,J.,Song,C.,Yue,Z.,Ding,S.:Facialvideo-basednon-contactstressrecognition utilizing multi-task learning with peak attention. IEEE Journal of Biomedical and Health Informatics28(9), 5335–5346 (2024)

    work page 2024

  43. [43]

    Sensors20(19) (2020)

    Zhang, H., Feng, L., Li, N., Jin, Z., Cao, L.: Video-based stress detection through deep learning. Sensors20(19) (2020)

    work page 2020