Analysing gait features via data-driven approaches

Abstract

Work in this thesis is about analysing two types of kinematics data representation: spatial representation and temporal representation. Spatial representation data is proposed to be silhouette moments data and temporal representation data is proposed to be angular displacements data. These data are analysed through a data-driven approach, which employs Principal Component Analysis (PCA) and Canonical Analysis (CA). PCA is a feature representation technique, which aims at reducing input data dimensionality without sacrificing the discriminative capability of the input data information; while CA is a feature discrimination technique, which aims at discriminating the input data for the best possible projection into the feature space. Before the input data are applied to the PCA and CA algorithm, they are pre processed in a cycle extraction procedure, which involves interpolation and resampling, to ensure the analysis is invariant to different start and end points of a gait cycle. Previous approaches in gait recognition research have depended upon heel-strike frames to determine a gait cycle. Thus, this cycle extraction procedure can relieve this dependability. Results on using the proposed features (angular displacements and silhouette moments) have shown potential and performance is comparable to other literatures. Angular displacements features achieved 98% classification and silhouette moments features achieved 91 % classification on a sample database of 10 subjects with 14 sequences each. Findings have shown that angular displacements data is a much better data representation than silhouette's moments.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata