Anomaly gait detection in children with autism spectrum disorder based on markerless-model approach / Nur Khalidah Zakaria

Zakaria, Nur Khalidah (2020) Anomaly gait detection in children with autism spectrum disorder based on markerless-model approach / Nur Khalidah Zakaria. PhD thesis, Universiti Teknologi MARA.

Abstract

Autism spectrum disorder or ASD is known as one of neuro-developmental disorder that consists of a range of conditions characterised by some degree of impairment and the symptom(s) can normally be seen at an early age. However, this lifelong condition has no known cure, but the condition can be reduced and recovered through therapies and treatments. At present, numerous studies have been done in investigating the behaviours among ASD children especially in their walking gait. Different approaches are used in order to highlight the gait features that may be associated with differentiating the walking gait of these children. Several studies experimented with gait features using different types of gait techniques. Among the gait techniques, there is limited literature existed for the markerless-based gait techniques in gait analysis for ASD. Therefore, in this thesis, anomaly gait detection in children with autism using the markerless-based approach is proposed. Firstly, the markerless-model approach features are extracted from 23 ASD subjects and 30 typically development (TD) subjects using Kinect Sensor. The originally extracted gait data is represented by the depth values (m) by 300 observations for each group. From these attributes, 10 sets of features are derived and divided into two categories namely Direct Joint Feature (DIR), and Reference Joint Feature (REF). Next, the principal component analysis (PCA) and linear discriminant analysis (LDA) techniques are used as feature extraction in order to determine the significant features, and the outputs of new features are labelled or known as DIR PCA-CPV, DIR PCA-Scree, DIR LDA and REF PCA-CPV, REF PCA-Scree and REF LDA features. Further, to evaluate the effectiveness of these new groups‘ features, four different classifiers namely the Support Vector Machine (SVM), Naïve Bayesian (NB), Artificial Neural Network (ANN) and Long-Short Term Memory (LSTM) are used to classify the gait samples from the different sets of features for the two groups of subjects as ASD or TD. Results attained showed that the NB classifier achieved the highest accuracy with 99.33% using DIR LDA specifically Set 1 that represents the full body and extracted by selecting all body points in x, y, and z axes. Additionally, this feature set is capable to correctly classify the ASD group with sensitivity of 99.66% and the TD group with specificity of 99.00%. In addition, based on the utilised of the intra-group and inter-group normalisation for the evaluation and validation of the hip, knee and ankle angles, the kinematic plots showed that the hip and the knee plots are similar for both techniques based on comparison of the markerless-based technique versus the marker-based technique. The ANOVA test supported the analysis that showed the insignificant means of differences for all angles. Hence, it is proven that the proposed markerless-based gait technique and the proposed gait feature are indeed apt to be used in classifying the gait of both ASD and normal children with LDA as the most suitable feature extraction and NB as the optimum classifier.

Metadata

Item Type: Thesis (PhD)
Creators:
Creators
Email / ID Num.
Zakaria, Nur Khalidah
2016200616
Contributors:
Contribution
Name
Email / ID Num.
Thesis advisor
Md Tahir, Nooritawati (Prof. Dr.)
UNSPECIFIED
Subjects: Q Science > QA Mathematics > Evolutionary programming (Computer science). Genetic algorithms
Divisions: Universiti Teknologi MARA, Shah Alam > Faculty of Electrical Engineering
Programme: Doctor of Philosophy (Electrical Engineering)
Keywords: Autism spectrum disorder; walking gait
Date: November 2020
URI: https://ir.uitm.edu.my/id/eprint/61067
Edit Item
Edit Item

Download

[thumbnail of 61067.pdf] Text
61067.pdf

Download (48kB)

Digital Copy

Digital (fulltext) is available at:

Physical Copy

Physical status and holdings:
Item Status:

ID Number

61067

Indexing

Statistic

Statistic details