Pentagon profile as image representation for human action interpretation task

Abstract

The emergence of various vision-based applications has led researchers to do research on human action recognition. Good feature extraction technique is one of the important factors in making an efficient action recognition system. In this work, we propose a technique which we called ‘pentagon profile’ to represent the image for the human action interpretation task. This profile is generated by connecting the points detected at the human head, hand and feet of a human silhouette image. These points when connected form a unique shape with five-sided polygons that prompt the name pentagon profile. Initial finding shows that this feature extraction technique gives promising result to represent the human activity especially in standing and squatting posture.