Virtual reality platforms with emotion recognition and animation capabilities for self-attachment technique
Author(s)
Polydorou, Neophytos
Type
Thesis
Abstract
This thesis aimed to enhance a novel self-administrated psychotherapeutic procedure called Self-Attachment Technique (SAT) using technological tools such as virtual reality (VR) and machine learning. The main question was how efficient and scalable SAT intervention can become using photorealistic VR avatars and automatic emotion recognition. Initially, inspired by a previous version, I developed a high-end platform for the Oculus Quest VR headset, which I evaluated through a small trial and showed that interactions with personalised virtual avatars were more effective and realistic than with generic avatars. However, VR headsets are costly and therefore, to make the procedure more scalable, I designed a low-end VR platform with personalised avatars that allow users to practice the SAT exercises on mobile devices with a cheap Google Cardboard. This platform was used to conduct the first 8-week non-clinical trial for the VR-based SAT intervention. The results showed significantly large effects on participants' wellbeing and self-compassion, as well as moderate effects on psychological capital. Most importantly, exploratory analysis revealed that these effects were much greater for participants who practised SAT with their avatars than those who used static photos. Similar significant improvements in wellbeing and self-compassion were obtained from smaller follow-up trials. In addition to the VR platforms, I further enhanced SAT by developing a machine learning model for facial emotion recognition, which achieved high accuracy on partially occluded faces due to the VR headset obstruction during SAT sessions. Overall, I successfully advanced tools for the efficient delivery of SAT and demonstrated for the first time that VR-based SAT can significantly enhance the wellbeing of healthy individuals.
Version
Open Access
Date Issued
2024-09-25
Date Awarded
01/02/2025
License URL
Advisor
Edalat, Abbas
Nicholls, Dasha
Sponsor
UK Research and Innovation
Grant Number
EP/S023283/1
Publisher Department
Department of Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)