Entrenamiento de una red neuronal Convolucional para la prueba Sit To Stand en rehabilitación
Loading...
Date
2026-04-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Tecnológica Centroamericana UNITEC
Abstract
El método tradicional para rastrear el movimiento del cuerpo humano fue el uso de sistemas de captura de movimiento basados en marcadores infrarrojos, una tecnología ampliamente utilizada para medir el rango de movimiento (ROM). Este sistema se consideró el "estándar de oro", ya que ofrecía alta precisión y fiabilidad; sin embargo, requería una alta inversión monetaria para poder emplearse.La captura de movimiento sin marcadores se presentó como una alternativa a la captura de movimiento basada en marcadores. Esta técnica resultó más asequible y fácil de usar, pero también demostró tener el potencial de ser más precisa.Con el concepto más claro, se presentó la idea de cómo este tipo de sistemas de captura de movimiento pueden impulsar la mejora del diagnóstico médico en rehabilitación, brindando asistencia de manera remota.Para hacer funcionar este proyecto, fue requerida la creación de una base de datos que contenía una serie de videos de personas ejecutando un ejercicio de rehabilitación llamado Sit To Stand (STS), el cual requiere que el paciente se levante y se siente de una silla sin apoyabrazos durante 30 segundos, realizando la mayor cantidad de repeticiones posibles. Esta prueba es esencial para medir la capacidad física de la persona.Uno de los objetivos de este estudio fue la aplicación de una red neuronal convolucional (CNN, por sus siglas en inglés), capaz de detectar las veces en que una persona se levanta y se sienta durante la ejecución de la prueba STS, apoyando significativamente el área de la telemedicina.Además, este enfoque resulta de gran utilidad para los centros de rehabilitación y promueve la inclusión de personas con discapacidad, ya sea por enfermedad o por causas genéticas, facilitando su acceso a procesos de rehabilitación. Gracias a esta CNN, los pacientes tienen la oportunidad de recibir atención del personal médico sin necesidad de salir de la comodidad de su hogar.
The traditional method for tracking human body movement involved the use of infrared marker-based motion capture systems, a widely used technology for measuring Range of Motion (ROM). This system was considered the "gold standard" as it offered high precision and reliability; however, it required a significant monetary investment to be implemented.Markerless motion capture emerged as an alternative to marker-based motion capture. This technique proved to be more affordable and user-friendly, while also demonstrating the potential to be more accurate.With a clearer understanding of the concept, the idea was introduced of how such motion capture systems could enhance medical diagnosis in rehabilitation by providing remote assistance.To operationalize this project, the creation of a database was required, containing a series of videos of individuals performing a rehabilitation exercise called the 30-second Sit-to-Stand (STS). This exercise requires the patient to stand up and sit down from a chair without armrests for 30 seconds, aiming to complete as many repetitions as possible. This test is essential for measuring the physical capacity of the individual.One of the objectives of this study was the application of a Convolutional Neural Network (CNN), capable of detecting the number of times a person stands up and sits down during the execution of the STS test, strongly supporting the field of telemedicine.Additionally, this approach is highly beneficial for rehabilitation centers and promotes accessibility for individuals with disabilities, whether due to illness or genetic conditions. Thanks to this CNN, patients have the opportunity to receive medical attention without the need to leave the comfort of their homes.
The traditional method for tracking human body movement involved the use of infrared marker-based motion capture systems, a widely used technology for measuring Range of Motion (ROM). This system was considered the "gold standard" as it offered high precision and reliability; however, it required a significant monetary investment to be implemented.Markerless motion capture emerged as an alternative to marker-based motion capture. This technique proved to be more affordable and user-friendly, while also demonstrating the potential to be more accurate.With a clearer understanding of the concept, the idea was introduced of how such motion capture systems could enhance medical diagnosis in rehabilitation by providing remote assistance.To operationalize this project, the creation of a database was required, containing a series of videos of individuals performing a rehabilitation exercise called the 30-second Sit-to-Stand (STS). This exercise requires the patient to stand up and sit down from a chair without armrests for 30 seconds, aiming to complete as many repetitions as possible. This test is essential for measuring the physical capacity of the individual.One of the objectives of this study was the application of a Convolutional Neural Network (CNN), capable of detecting the number of times a person stands up and sits down during the execution of the STS test, strongly supporting the field of telemedicine.Additionally, this approach is highly beneficial for rehabilitation centers and promotes accessibility for individuals with disabilities, whether due to illness or genetic conditions. Thanks to this CNN, patients have the opportunity to receive medical attention without the need to leave the comfort of their homes.
Keywords
Computer Vision, Convolucional Neuronal Network, Deep Learning, Kinovea, Sit To Stand