Human Posture Detection on Lightweight DCNN and SVM in a Digitalized Healthcare System

Abstract

The domains of Human Posture (HP) and artificial intelligence (AI) have placed significant emphasis on Human Posture Classification (HPC). By recognizing the standing, sitting, and walking positions of aged people, HPC could effectively monitor their health condition and ensure their well-being. Detection of various postures is difficult owing to the need for enough datasets and a classification framework. MobileN et is a lightweight Deep Convolution Neural Network (DCNN) with a better recognition rate and fewer parameters. To further increase the generalization ability while reducing the number of models hyperparameters, techniques such as regularization, Transfer Learning (TL) and neural architecture search can be employed. To address the issue of a shortage of annotated data and improve approach effectiveness, TL and Data Augmentation (DA) are utilized for MobileNet and Xception in this study. The Support Vector Machine (SVM) model boosts performance instead of the final Fully Connected Layer (FCL). To facilitate the effectiveness of our approach, TL and DA are both used for SVM. The investigation associated the efficacy of the suggested approach with that of other cutting-edge image classification techniques, including ResNet50V2, Inception V3,and DenseN et201, and found that the presented methodology is better. Our recommended method captures temporal and depth characteristics from the image separately and incorporates them into classification computations. The suggested approach founded on MobileN et hybridized with SVM attains the top performance with 92.12% test accuracy (Acc), 95 % area under the curve (AUC), 92% recall (Rec), 93% precision (Prec), 92 % F1 score and a computational period of 3974secs.