EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Comparative Study of Three AI Approaches for Human Movement Classification in Real-Time Physiotherapy and Action Recognition

Eldhose PU, R Gunasundari

Journal of Intelligent Decision Making and Information Science · 02.09.2026 · 3 (10s) · 293-308

DOI: 10.17762/jidmis.v3.2718 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 02.09.2026. Verfasst von Eldhose PU und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid growth of telemedicine, digital healthcare ecosystems, and intelligent rehabilitation technologies has significantly increased the demand for scalable, automated, and reliable physiotherapy monitoring systems. Traditional physiotherapy assessment relies heavily on in-person clinical supervision, manual observation, and subjective feedback, creating challenges related to accessibility, cost, geographical limitations, and therapist availability. These limitations became particularly evident during the COVID-19 pandemic, which accelerated the adoption of remote healthcare and tele-rehabilitation services. Consequently, there is a growing need for intelligent systems capable of monitoring exercise performance, evaluating movement quality, and providing real-time corrective feedback without requiring specialized sensors or continuous expert supervision. This study presents a comparative analysis of three artificial intelligence-based human movement classification frameworks: PoseActionDetector, CNN-Based Physiotherapy System, and EfficientNetB0-LSTM Hybrid Architecture. These approaches represent three distinct paradigms of movement assessment, namely rule-based reasoning, lightweight convolutional learning, and hybrid deep-learning-based temporal modeling. The proposed comparative framework integrates pose estimation, skeletal landmark analysis, convolutional neural networks, temporal sequence learning, attention mechanisms, and mobile deployment optimization techniques for accurate action recognition and physiotherapy monitoring.The PoseActionDetector employs geometric reasoning and temporal logic on pose landmarks extracted through MediaPipe, providing computational efficiency and explainability suitable for low-resource environments. The CNN-based Physiotherapy System combines BlazePose skeletal extraction, adaptive preprocessing techniques, and lightweight convolutional learning to support home-based rehabilitation assessment under varying environmental conditions. The EfficientNetB0-LSTM Hybrid Architecture integrates deep spatial feature extraction with temporal sequence modeling and attention mechanisms, enabling robust movement classification and exercise form evaluation in complex real-world scenarios. Experimental evaluation demonstrates that the EfficientNetB0-LSTM Hybrid achieves superior performance, reaching 97.2% multi-class classification accuracy and 100% binary form-validation accuracy while maintaining real-time operation on commodity smartphones. The PoseActionDetector achieves high interpretability and low computational overhead with up to 95% accuracy in controlled environments, whereas the CNN-based framework provides a balanced trade-off between accuracy, robustness, and deployment efficiency. The findings emphasize the importance of feedback mechanisms. The proposed comparative analysis contributes toward the physiotherapy systems capable of transforming future telehealth and rehabilitationinfrastructures.

Zitieren

APA

PU, E., & Gunasundari, R. (2026). Comparative Study of Three AI Approaches for Human Movement Classification in Real-Time Physiotherapy and Action Recognition. Journal of Intelligent Decision Making and Information Science, 3(10s), 293-308. https://doi.org/10.17762/jidmis.v3.2718

Vancouver

PU E, Gunasundari R. Comparative Study of Three AI Approaches for Human Movement Classification in Real-Time Physiotherapy and Action Recognition. Journal of Intelligent Decision Making and Information Science. 2026;3(10s):293-308. doi:10.17762/jidmis.v3.2718

BibTeX
@article{PU2026,
  author = {PU, Eldhose and Gunasundari, R},
  title = {Comparative Study of Three AI Approaches for Human Movement Classification in Real-Time Physiotherapy and Action Recognition},
  journal = {Journal of Intelligent Decision Making and Information Science},
  year = {2026},
  volume = {3},
  number = {10s},
  pages = {293--308},
  doi = {10.17762/jidmis.v3.2718}
}
Kostenlos starten →