Physical rehabilitation techniques during the treatment of clinical pathologies, is one of the most challenging areas for the medical structure and for patients and families. In continental countries like Brazil, the need of remote monitoring of this type of treatment is necessary and important. However, equipments and medical follow-up during exercises still have high costs.With the improvement of Computer Vision and Machine Learning techniques, some computational less expensive alternatives have been proposed in the literature. However, the monitoring of patients during physical rehabilitation exercises with the help of artificial intelligence by a health professional, especially from the capture of visual signals, is still a challenge poorly explored in the scientific-technological literature. This work presents the development of a system based on a RBG camera for machine learning training to track people joints in real time, to optimize the results of physical therapy sessions. Two modules were created using the concept of modular neural networks: a module for detection and another for measurement. The Detection module detects the different types of rehabilitative physical exercises, while the Measure Module validates if the exercise is being done correctly and, if not, suggests a correction. In all, only three types of exercises were considered: Squat, knee flexion and hip extension. After analyzing the angles of the joints of each of the exercises, it was found that only four angles would be necessary to describe the exercises: Angle of the armpits, hip, knee and lower-limbs (between the legs). To perform the tests, three databases were generated: Base-Original, Original-Oscilação and Oscilação-Falhas. The Base-Original is made up of only four original angles captured from the people performing the exercises. The Original-Oscilação, is composed by the Base-Original, but with small variations in its four angles. The Oscilação-Falhas base is composed of the Base-Original, but with large oscillations, in order to make the exercise incorrect, but not mischaracterizing it. Each module has its own back-propagation neural network. To find out which are the best models to solve this type of problem, nine architectures were proposed for the Detection module and another nine for the Measure module. For the Detection module, architectures from 1-1 to 9-1 (categories 1 through 9, with the number after the dash denoted as the module type, in this case 1 being Detection) were generated, trained with Base-Original and Original-Oscilação. The same was done for the Measurement module, other architectures from 1-2 to 9-2 (categories 1 through 9, with the number after the dash denoted as the module type, in this case 2 being Measure) were generated, trained with the Oscilação-Falhas database.After generating the results, the 9-1 and 9-2 architectures had the best results. Both architectures achieved more than 90% accuracy, both in the recognition of exercises and in their validation.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.