The central aim of this seminal research work is to find the main features that mostly contribute to a gait signature. Principal Components Analysis is used to reduce the dimensionality of the data, which was obtained mainly via labelling. Canonical Analysis was then used to separate the projected data into predefined classes. K-nearest method is then applied to classify testing data in the canonical space. A Divide and Merge algorithm is derived to search for the best features that offer higher level of discriminability for gait signature, it is revealed that it is possible to achieve 100% CCR using only a set of 50 features which is a trade-off between noise and poor discriminability. The application of Fourier Frequencies Analysis revealed that the combination of the product of the magnitude and phase for the thigh and lower leg offer a discriminability capability to a gait identification system. Moreover, the Algorithm is extended to use of both selection criteria i.e. Correct Classification Rate and the Distribution J values, a better results is obtained using a set of only 20 features.

This entry was posted on Tuesday, December 20th, 2005 at 6:07 pm and is filed under Research. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.