Walking gestures recognition based on a novel symbolic representation

Abstract

This study presents a new method for walking gesture representation and recognition based on a symbolic representation. The symbolic representation is used to represent images which can be decomposed into a number of time series and the distance between time series can be characterized based on the distances between symbols. In this work, the distances between symbols are defined according to the average value of each segment rather than the distance calculated based on Gaussian distribution as used in traditional symbolic representations. The proposed method is applied to a short template video containing a number of walking steps (gestures). In the case studies we consider a database containing 104 test videos, taken from 101 people, and our objective is to identify whether a person in the testing video is the same person as in the template video (i.e. the training video), and the identification accuracy of the method is around 98%.