Extracting unique patterns from human actions

Abstract

Human walks, runs, dances and left behind interesting information on their actions. This paper presents how chaotic dynamics help to interpret and classify human actions. The trajectories of two legs are extracted during a motion such as walk. These trajectories of foot points are collected from an artificial human video arrangement. Each dimension of trajectory represents a time series. The phase space for each time series is reconstructed using appropriate time delay and dimension. The plot exhibited a characteristic trajectory representing the regularity of the time series. Analysis of time series obtained in human with three different motions revealed that the trajectory behaves in such a way that the time series is governed with a deterministic rule. Unique patterns are observed for a particular motion. This can be revealed from the phase space and self organizing map (SOM) network. The motions (walk, run, and jump) can be categorized in terms of different shape of phase space and output of the SOM network. The results are validated with correlation dimension. These representations are very useful in classifying the human motions.