Fuzzy transfer learning in human activity recognition.

Adama, DOA ORCID: https://orcid.org/0000-0002-2650-857X, 2020. Fuzzy transfer learning in human activity recognition. PhD, Nottingham Trent University.

Preview

Text __Opel.ads.ntu.ac.uk_IRep-PGR$_2020 Theses and deposit agreement forms_CST_SST_ADAMA_David_David_Adama_2020.pdf - Published version Download (11MB) | Preview

Abstract

Assisted living environments are incorporated with different technological solutions to improve the quality of life and well-being. In recent years, there has been a growing interest in the research community on how to develop evolving solutions to aid assisted living. Different techniques have been studied to address the need for technological systems which are intelligent enough to evolve their knowledge to solve tasks which have not been previously encountered. One such approach is Transfer Learning (TL), for example, between humans and robots.

Humans excel at dealing with everyday activities, learning and adapting to different activities. This comprises different complex techniques which enable the lifelong learning process from observation of our environment. To obtain similar learning in assistive agents, TL is needed. The aim of the research reported in this thesis is to address the challenge associated with learning and reuse of knowledge by assistive agents in an Ambient Assisted Living (AAL) environment. In this thesis, a novel approach to transfer learning of human activities through the combination of three methods; TL, Fuzzy Systems (FS) and Human Activity Recognition (HAR) is presented. Through the incorporation of FS into the proposed approach, uncertainty that is evident in the dynamic nature of human activities are embedded into the learning model.

This research is focused on applications in assistive robotics. This is with a purpose of enabling assistive robots in AAL environments to acquire knowledge of such activities as are performed by humans. To achieve this, an extensive investigation into existing learning methods applied in human activities is conducted. The investigation encompasses current state-of-the-art of TL approaches employed in skill transfer across different but contextually related activities.

To address the research questions identified in the thesis, the contributions of the methodology employed are in three main categories; 1) Firstly, a novel framework for human activity learning from information observed. Experiments are conducted on selected human activities to acquire enough information for building the framework. From the acquired information, relevant features extracted are used in a learning model to recognise different activities. 2) Secondly, the sequence of occurrence(s) of tasks in an activity needs to be considered in the learning process. Therefore, in this research, a novel technique for adaptive learning of activity sequences from acquired information is developed. 3) Finally, from the sequence obtained, a novel technique for transfer of human activity across heterogeneous feature space existing between a human and an assistive robot is developed. These categories form the basis of the TL framework modelled in this research.

The framework proposed is applied to TL of human activity from data generated experimentally and benchmark datasets of various classes of human activities. The results presented in this thesis show that exploring the process of human activity learning is an important aspect in the TL framework. The features extracted sufficiently distinguish relevant patterns for each activity. Also, the results demonstrate the ability of the methodology to learn and predict human actions with a high degree of certainty. This encourages the use of TL in assisted living environments and other applications. This and many more applications of TL in technology would be a potential driver of the next revolution in artificial intelligence.

Item Type:	Thesis
Creators:	Adama, D.O.A.
Date:	May 2020
Divisions:	Schools > School of Science and Technology
Record created by:	Jeremy Silvester
Date Added:	23 Jul 2020 14:28
Last Modified:	31 May 2021 15:18
URI:	https://irep.ntu.ac.uk/id/eprint/40268

Actions (login required)

Edit View

Statistics

Views

Downloads