Real-Time Posture Identification System for Wheelchair Users Preventing the Generation of Pressure Ulcers

Abstract

Prevention is key to avoid pressure ulcer generation in people with mobility restrictions. In recent years, preventive medicine has focused on posture control by considering people who frequently have the same position for too long, such as wheelchair users. Optical fiber sensors have gained recognition for their applications in biomedical engineering; however, approaches to assistive devices, such as wheelchairs, have been relatively unexplored. This study proposes a polymeric-optical-fiber (POF) sensing system based on machine learning (ML) for human posture recognition in an electrical wheelchair-based human machine interface (HMI). The ML-based model was used to classify time- and frequency-domain features obtained from a matrix of POF-based pressure sensors and 24 photodetectors during the execution of eight body postures. In an offline stage, multiclassification was conducted using k-nearest neighbors (KNN), decision tree, extra tree classifier (ETC), and random forest, where the best performance, in terms of accuracy (ACC), was obtained through the use of ETC (94%). Hence, this classifier was implemented in real-time, where the wheelchair-based HMI achieved a CPU time of approximately 117 ms, and an ACC higher than 96%, outperforming the metrics previously reported in the literature. We believe that this study contributes to the development of smart assistive systems that integrate ML and soft sensors to recognize body postures in an HMI, which is a promising approach for preventing the generation of pressure ulcers in wheelchair users.