Comparing Posture Classification: A Human Lying Posture Pressure-Map Dataset
Loading...
Date
2025-04-11
Authors
Husák, Michal
Mihálik, Ondrej
Arm, Jakub
Mesárošová, Michaela
Kaczmarczyk, Václav
Bradáč, Zdeněk
0000-0002-8743-6776Advisor
Referee
Mark
Journal Title
Journal ISSN
Volume Title
Publisher
IEEE
Altmetrics
Abstract
We discuss methods and algorithms for classifying the posture of a patient in their bed. The actual classification tasks are performed with a measurement chain including an in-house designed pressure mattress, a matrix of 30×11 sensing spots, a data concentrator, and a cloud-based service. Utilizing a survey of open-source datasets that facilitate such classifying operations, we designed a relevant experiment. A Human Lying Posture Pressure-Map dataset (HLPPDat) formed during the research is publicly available. Involving 20 subjects in 64 defined postures, the classification output was separated into four basic groups included prone posture. The data enabled us to analyze multiple classification methods developed with the state-of-the-art concepts of Machine Learning (ML), sparse representation, and artificial intelligence represented by Transfer Learning (TL). The analysis included both data measured and data experimentally corrupted in a manner that would most probably occur due to a partial measurement error. Regarding the techniques and options tested, feature extraction via the Histogram of Oriented Gradient (HOG) and the K-Nearest Neighbors (KNN) tools appeared to be the most beneficial, yielding an accuracy of over 99.5% in the leave-one-subject-out crossvalidation. The research confirmed that an accurate classification is feasible at a matrix sensor resolution markedly lower than the limits regularly presented in the literature. The system allows monitoring how long a bedridden person has remained in the same posture, and thus it has a potential to help prevent decubitus in both hospitals and the home care.
Description
Citation
Document type
Peer-reviewed
Document version
Published version
Date of access to the full text
Language of document
en