Clasificación de posturas de equilibrio del Sppb utilizando redes reuronales y visión por computadora
Loading...
Date
2026-04-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Tecnológica Centroamericana UNITEC
Abstract
El envejecimiento poblacional conlleva un aumento del riesgo de fragilidad y pérdida de equilibrio en adultos mayores, lo que incrementa la probabilidad de caídas y afecta su autonomía.
Para abordar este problema, este estudio desarrolla un modelo basado en visión por computadora y redes neuronales convolucionales (CNN) para la clasificación de posturas de equilibrio en la prueba SPPB.
El sistema emplea Roboflow para entrenar el modelo de aprendizaje profundo con imágenes segmentadas de videos. Se han etiquetado posturas clave del SPPB (pies juntos, semi-tándem y tándem) para permitir la identificación automática.
El modelo será validado mediante métricas de desempeño como precisión, recall y mAP, garantizando su fiabilidad para futuras aplicaciones en la evaluación del equilibrio.
Este enfoque elimina la necesidad de sensores físicos y proporciona una alternativa accesible para el análisis postural en entornos clínicos y de investigación.
The aging population leads to an increased risk of frailty and loss of balance in older adults, which raises the likelihood of falls and affects their autonomy. To address this issue, this study develops a model based on computer vision and Convolutional Neural Networks (CNNs) for the classification and tracking of balance postures in the Short Physical Performance Battery (SPPB). The system uses Roboflow to train the deep learning model with segmented images from videos. Key postures from the SPPB (feet together, semi-tandem, and tandem) have been labeled to allow for automatic identification. The model will be validated using performance metrics such as accuracy, recall, and mAP, ensuring its reliability for future applications in balance assessment. This approach eliminates the need for physical sensors and provides an accessible alternative for postural analysis in clinical and research settings.
The aging population leads to an increased risk of frailty and loss of balance in older adults, which raises the likelihood of falls and affects their autonomy. To address this issue, this study develops a model based on computer vision and Convolutional Neural Networks (CNNs) for the classification and tracking of balance postures in the Short Physical Performance Battery (SPPB). The system uses Roboflow to train the deep learning model with segmented images from videos. Key postures from the SPPB (feet together, semi-tandem, and tandem) have been labeled to allow for automatic identification. The model will be validated using performance metrics such as accuracy, recall, and mAP, ensuring its reliability for future applications in balance assessment. This approach eliminates the need for physical sensors and provides an accessible alternative for postural analysis in clinical and research settings.
Keywords
Análisis Postural, Aprendizaje Profundo, Clasificación de Posturas, Equilibrio, Redes Neuronales Convolucionales, RoboFlow, Seguimiento de Posturas, Visión por Computadora, SPPB