Repository logo
 

Sensor-Based Human Activity Recognition in a Multi-Resident Smart Home Environment

Files

Thesis(4.58 MB)

Date

2021

Authors

Supervisor

Yonchareon, Sira
Lai, Edmund

Item type

Thesis

Degree name

Doctor of Philosophy

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

Human Activity Recognition (HAR) uses various types of sensors to collect data and recognise human motions. HAR can prevent potentially unsafe movements and behaviours such as falling or forgetting to take a medication. Many practical applications of HAR have emerged in recent years, such as ambient assisted living (AAL) systems and smart home monitoring. The vast majority of current behaviour recognition solutions in smart home environments are designed for single residents. However, in real life, living environments are occupied not only by the target individual, but also by their family and/or guests, which complicates the task of activity recognition. Therefore, multi-resident activity recognition is demanded in a smart home environment. This research addresses three sub-problems of the multi-resident activity recognition problem: segmentation, classification, and online learning. To solve the segmentation problem, we first propose a novel hybrid fuzzy C-means segmentation method based on change point detection (CPD) for sensor events, which improves the performance of multi-resident activity recognition. Next, we propose a new Transformer with Bidirectional Gated Recurrent Unit (Bi-GRU) deep-learning method called TRANS-BiGRU, which efficiently learns and recognises the complex activities of multi residents. Finally, we propose a novel Locally-weighted Ensemble Detection-based Adaptive Random Forest Classifier (LED-ARF) for online analysis of multi-resident identification. After comprehensive experiments, we found that our proposed algorithm effectively solves the three problems of multi-resident action recognition.

Description

Keywords

Multi-resident activity recognition, Activity sequence segmentation, Deep learning, Online learning

Source

DOI

Publisher's version

Rights statement

Permanent link

https://hdl.handle.net/10292/14644

Collections

Doctoral Theses