| Early Detection of Postural Abnormalities Using Computer Vision and Artificial Intelligence: A Narrative Review |
| Paper ID : 1129-SPORTCONGRESS (R1) |
| Authors |
|
Mohamad ali Biabangard *, Saeid Azadbakht, Ali asghar Maleki Department of Sport Injuries and Biomechanics, Faculty of Sport Sciences and Health, University of Tehran |
| Abstract |
| Introduction: Postural abnormalities such as scoliosis, kyphosis, and forward head posture contribute to musculoskeletal pain, impaired performance, and higher injury risk. Conventional screening methods are often slow, costly, and insufficient for early detection. Advances in computer vision and machine learning now allow accurate and automated posture assessment. This study reviews recent applications of artificial intelligence in posture evaluation and its role in developing corrective exercise programs. Methods: This study was conducted as a narrative review was searched in, and between 2020 and 2025 using keywords such as “computer vision,” “posture assessment,” “machine learning,” and “corrective exercise.” Selected sources were evaluated based on their conceptual relevance and scientific quality and were analyzed narratively. Results: The reviewed studies demonstrated that computer vision algorithms especially convolutional neural networks (CNNs) and human pose estimation models achieved classification accuracies ranging from 85% to 95% in controlled settings. Most approaches relied on 2D RGB images or videos captured by standard cameras, while a few employed depth sensors or radar-based systems. These models effectively identified postural deviations such as scoliosis, pelvic tilt, and knee valgus without requiring specialized clinical equipment. Some systems also offered real-time feedback to guide corrective exercises, indicating their potential for integration into tele-rehabilitation platforms. Conclusion: AI-based posture assessment technologies hold substantial promise as efficient, scalable, and low-cost tools for early detection of musculoskeletal abnormalities and guiding individualized corrective exercise programs. However, translating these systems from controlled laboratory environments to real-world clinical or community settings requires overcoming challenges such as dataset diversity, robustness to environmental changes, and data privacy. Future research should focus on developing standardized and publicly accessible datasets, validating models across diverse populations, and integrating algorithms into user-friendly tele-rehabilitation platforms. Ultimately, these advances could enable accessible, continuous, and objective posture monitoring for large populations, reducing healthcare burdens and improving long-term musculoskeletal health. |
| Keywords |
| Computer Vision; Posture Assessment; Machine Learning; Corrective Exercise; Early Detection |
| Status: Abstract Accepted (Poster Presentation) |
