A novel Approach to Human Gait Recognition using possible Speed Invariant features

In this paper a new area based technique is proposed for deriving gait signatures by decomposing the human body into three independent structural segments such as head node, arm swing and leg swing areas. Initially, all the feature points are represented as the sides of an n-sided polygon for calculating the area of each region. This technique induces surplus noise in the feature points which is in turn reflected in the human identification problem. This drawback inspires us to compute the area of each region by constructing a convex hull of the feature points in order to obtain certain key speed invariant features. Classification results demonstrate the ability of proposed feature extraction techniques using Bayes’ classifier, distance metrics, and the proposed polynomial based distance metric. The performance analysis of various classifiers has been evaluated using Receiver Operating Characteristics (ROC) curve and the Cumulative Match Characteristics Curve (CMC) after performing N-fold cross validation technique.