Title: Deep learning of human posture image classification using convolutional neural networks

Authors: Aaron Rasheed Rababaah

Addresses: College of Engineering and Applied Sciences, American University of Kuwait (AUK), Salmiya, 13034, State of Kuwait

Abstract: In this paper a study of deep learning applied to human posture image classification using convolutional neural networks (CNNs) is presented. Typical computer vision workflow includes in the early stages: data conditioning, feature extraction, dimensionality reduction/feature selection whereas, in CNNs, these stages are eliminated which provides a big advantage of automatic feature extraction. In this work, CNNs are applied to human posture classification. Collected human postures included standing with five different variations, sitting with two different variations, bending and sleeping with two different variations. More than 6000 samples were collected for training and validation. Several independent experiments were conducted each of which has a different number of filters/kernels ranging within [1, 32]. The results of the experimental work showed that number of features influenced the classification accuracy significantly as the lowest CNN model produced 91.76% and the highest model produced 98.57% classification accuracy.

Keywords: human posture classification; image processing; machine vision; deep learning; CNNs; convolutional neural networks.

DOI: 10.1504/IJCSM.2022.124708

International Journal of Computing Science and Mathematics, 2022 Vol.15 No.3, pp.273 - 288

Received: 05 Dec 2020
Accepted: 23 Apr 2021
Published online: 08 Aug 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article