| Summary: | The study of human walking has been of interest to researchers in different disciplines. Their interest included but was not limited to modelling of the human body, perceiving actions, analysing gait, etc. The data collected includes eighty normal and thirty pathological subjects. The work presented comprises tracking of the markers, analysis of the motion trajectories, extraction of salient features and recognition of gait signatures using a vector quantiser and a clustering mechanism of different groups of subjects - normal and pathological - through interpreting the knowledge obtained from the clusters. A computer program has been implemented to illustrate the ideas of this work. Radial basis functions neural networks are used for tracking the makers, predicting their positions in case of occlusions as well as interframe to obtain a complete smooth trajectory for the motion of the joints in 3D. The analysis algorithm is based on combining the wavelet transform for feature extraction and Kohonen self-organising map (SOM) vector quantiser for classification of the walking patterns. Rules are then extracted from the SOM after self-organisation to determine the salient features characterising each cluster as well differentiating it from others. The approach is demonstrated by its application to kinematic gait data for both normal and pathological subjects. It is shown and experimentally verified that salient features do exist within the internal structure of the kinematic data from which diagnostic signatures are elicited. Existence of such features could be used by clinicians in the orthopaedic field where the gait disease signatures would well mean improved assessment of gait and treatment and possibly early detection of some locomotion impairment.
|
|---|
