A system for the visual detection and analysis of obsessive compulsive disorder

Cameron, C 2018, A system for the visual detection and analysis of obsessive compulsive disorder, Doctor of Philosophy (PhD), Science, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Cameron.pdf Thesis application/pdf 1.67MB
Title A system for the visual detection and analysis of obsessive compulsive disorder
Author(s) Cameron, C
Year 2018
Abstract Computer vision is a burgeoning field that lends itself to a diverse range of challenging problems. Recent advances in computing power and algorithmic sophistication have prompted a renaissance in the literature of this field, as previously computationally expensive applications have come to the fore. As a result, researchers have begun applying computer vision techniques especially prominently to the analysis of human actions, in an increasingly advanced manner. Chief among the potential applications of such human action analyses are: human surveillance, crowd analysis, gait analysis and health informatics. Even more recently, researchers have begun to realise the potential of computer vision techniques, occasionally in conjunction with other computational approaches, to enhance the quality of life for people living with mental illness. Much of this research has focused on enhancing the existing, traditionally psychiatric, treatment plans for such individuals. Conventionally, these treatment plans have involved a mental health professional taking a face-to-face approach and relying significantly on subjective feedback from the individual, regarding their current condition and progress. However, recent computational methods have focused on augmenting such approaches with objective, e.g. visual, monitoring and feedback on an individual's condition over time. Of these approaches, most have focused on depression, bipolar disorder, dementia, or some form of anxiety. However, none of the approaches described in the literature has been aimed directly at addressing the issues inherent to patients with Obsessive Compulsive Disorder. Motivated by this, the proposed thesis comprises the design and implementation of a system that is capable of detecting and analysing the compulsive behaviours exhibited by individuals with Obsessive Compulsive Disorder. This is accomplished with the aim of assisting mental health professionals in their treatment of such patients. We achieved the aforementioned via a three-pronged approach, which is represented by the three core chapters of this thesis. Firstly, we created a system for the detection of general repetitive (compulsive) behaviours indicative of Obsessive Compulsive Disorder. This was achieved via the use of a combination of optical flow detection and thresholding, an image matching algorithm, and a set of repetition parameters.

Via this approach, we achieved good results across a set of three tested videos. Secondly, we proposed a system capable of classifying behaviour as either compulsive or non-compulsive based on the differences in the repetition intensity patterns across a set of behavioural examples. We achieved this via a form of motion history image, which we call a 'Temporal Motion Heat Map' (TMHM). We produced one such heat map per behavioural example and then reduced its dimensionality using histogram-based pixel intensity frequencies, before feeding the result into a Neural Network. This approach achieved a high classification accuracy on the set of 40 tested behavioural examples, thus demonstrating its ability to accurately differentiate between compulsive and non-compulsive behaviours, as compared to a set of existing approaches. Finally, we built a system that is capable of categorising different types of behaviour, both compulsive and non-compulsive, and then assessing them for relative approximate anxiety levels over time. We achieve this using a combination of Speeded-Up Robust Features (SURF) descriptors for behaviour classification and statistical measures for determining the relative anxiety of a given compulsion. This system is also able to achieve a good accuracy when compared with other approaches.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Science
Subjects Computer Vision
Pattern Recognition and Data Mining
Image Processing
Keyword(s) Computer vision
Obsessive Compulsive Disorder
Machine learning
Repetitive behaviour
Motion analysis
Version Filter Type
Access Statistics: 48 Abstract Views, 139 File Downloads  -  Detailed Statistics
Created: Fri, 30 Nov 2018, 08:50:47 EST by Keely Chapman
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us