| Title |
Using skeleton detection to identify stress-induced changes in human movement patterns |
| Translation of Title |
Skeleto aptikimo metodų taikymas nustatant streso sukeltus žmogaus judesių pokyčius. |
| Authors |
Klumbys, Marius |
| Full Text |
|
| Pages |
53 |
| Keywords [eng] |
stress detection ; body pose estimation ; facial expression analysis ; multimodal feature fusion ; supervised learning |
| Abstract [eng] |
This study investigates video-based stress detection using body pose and facial expression features. A unified feature extraction pipeline gathers posture and facial data from video and aggregates them into fixed windows. This research proposes and systematically evaluates a video-based stress recognition framework that fuses body pose and facial expression features and benchmarks classical machine learning and sequence models. Three models: HistGradientBoosting, SVM with RBF kernel and BiLSTM are evaluated on two datasets: SWELL-KW and StressID. On SWELL-KW dataset, the BiLSTM achieved the highest balanced accuracy (~0.60), while on StressID dataset, HistGradientBoosting reached the best balanced accuracy (~0.74). The results indicate that dataset characteristics strongly influence model performance, and while video-only features provide a meaningful stress signal, further methodological improvements are needed to enhance detection accuracy. |
| Dissertation Institution |
Kauno technologijos universitetas. |
| Type |
Master thesis |
| Language |
English |
| Publication date |
2026 |
