Stress assessment in industrial and assembly environments using facial thermal imaging: A systematic bibliometric analysis

This thesis provides a comprehensive systematic and content-based review of seventy-one studies on facial thermal imaging for stress and workload assessment, with a particular focus on its applicability in manufacturing and assembly environments. The analysis examines publication trends, experimental paradigms, sensor characteristics, ROI selection strategies, thermal feature extraction methods, and machine learning and deep learning approaches, highlighting how the field has evolved from early psychophysiological validation studies to more sophisticated multimodal and application-oriented research. The findings show that thermal responses are strongly influenced by environmental and contextual factors, with controlled laboratory settings producing more consistent patterns than real-world or semi-naturalistic environments. Higher-resolution cameras, dynamic ROI tracking, derivative-based thermal features, and multimodal integration with ECG, HRV, EDA, or behavioral cues consistently enhanced detection accuracy. Despite this progress, several limitations remain, including motion artifacts, occlusion, environmental instability, limited worker diversity, and the absence of standardized multimodal datasets, which restrict the feasibility of deploying thermal imaging in industrial settings. The review contributes to a clearer theoretical understanding of autonomic thermal responses and provides practical implications for ergonomics, worker monitoring, cognitive-aware task design, and Industry 5.0 systems. The thesis concludes with recommendations for future research, emphasizing the need for real-world industrial validation, improved tracking and environmental compensation methods, larger and more diverse participant samples, and the development of open-access multimodal datasets to support robust and generalizable stress assessment.