An unsupervised statistical representation learning method for human activity recognition

With the evolution of smart devices like smartphones, smartwatches, and other wearable devices, motion sensors have been integrated into these devices to collect data and analyze human activities. Consequently, sensor-based Human Activity Recognition (HAR) has emerged as a significant research area...

Full description

Saved in:
Bibliographic Details
Published in:Signal, image and video processing Vol. 18; no. 10; pp. 7041 - 7052
Main Authors: Abdi, Mohammad Foad, BabaAli, Bagher, Momeni, Saleh
Format: Journal Article
Language:English
Published: London Springer London 01.09.2024
Springer Nature B.V
Subjects:
Accuracy
Algorithms
Computer Imaging
Computer Science
Computing costs
Datasets
Decision trees
Deep learning
Engineering
Fourier transforms
Human activity recognition
Human motion
Image Processing and Computer Vision
Machine learning
Motion sensors
Multimedia Information Systems
Neural networks
Original Paper
Pattern Recognition and Graphics
Performance evaluation
Regression analysis
Representations
Sensors
Signal,Image and Speech Processing
Smartwatches
Support vector machines
Ubiquitous computing
Vision
Wearable technology
Feature embedding
Latent Dirichlet allocation (LDA)
Human activity recognition
Unsupervised representation learning
ISSN:1863-1703, 1863-1711
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With the evolution of smart devices like smartphones, smartwatches, and other wearable devices, motion sensors have been integrated into these devices to collect data and analyze human activities. Consequently, sensor-based Human Activity Recognition (HAR) has emerged as a significant research area in the fields of ubiquitous computing and wearable computing. This paper presents a novel approach that employs Latent Dirichlet Allocation (LDA) to extract meaningful representations from activity signals. The method involves transforming the activity signal, which is a sequence of samples, into a sequence of discrete symbols using vector quantization. Subsequently, LDA is utilized to embed the symbol sequence into a fixed-length representation vector. Finally, a classifier is employed to classify the obtained representation vector. The effectiveness of the proposed method is evaluated using the UNIMIB-SHAR dataset. Experimental results demonstrate its competitive performance in terms of accuracy and F1-score metrics when compared to existing methods. Moreover, our method boasts a more lightweight architecture and incurs lower computational costs compared to deep learning-based approaches. The findings of this study contribute to the advancement of HAR and hold practical implications for HAR systems.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1863-1703
1863-1711
DOI:10.1007/s11760-024-03374-z