Enhancing Elderly Population Health Supervision Through Posture Detection Using Deep Learning

EasyChair Preprint no. 11928

Abstract

In light of the growing aging population and limited healthcare resources, there's a need for technology that supports the independence of the elderly through remote monitoring, particularly in maintaining proper posture, which is crucial for health. Posture recognition, the assessment of how one holds their body, is challenging due to scarce data and the need for real-time analysis. To tackle this, a dataset with over 7600 images of yoga poses was compiled. Dance poses add complexity to posture recognition due to their dynamic and multimodal nature. While most studies have used traditional machine learning (ML) classifiers for posture detection, they fall short in accuracy. This study introduces a novel hybrid approach that combines ML techniques—K-Nearest Neighbor (K-NN), Support Vector Machine (SVM), Naive Bayes, Random Forest, Logistic Regression, Decision Tree, Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis—with deep learning (DL) models like 1D and 2D Convolutional Neural Networks (CNNs), LSTM, and bidirectional LSTM. This hybrid method leverages the strengths of both ML and DL to improve prediction accuracy, achieving over 98% on a recognized benchmark dataset.

Keyphrases: 1D-CNN, 2D CNN, LSTM, Posture detection system, posture recognition

@Booklet{EasyChair:11928,
  author = {Shubhashish Jena and Gunamani Jena and Subhashree Jena},
  title = {Enhancing Elderly Population Health Supervision Through Posture Detection Using Deep Learning},
  howpublished = {EasyChair Preprint no. 11928},

  year = {EasyChair, 2024}}