Maximum intensity block code for action recognition in video using tree-based classifiers

Abstract

Human action recognition is a broad research area in computer vision community. Human actions are identified by connotation of body movements. In this paper, an action recognition approach based on maximum motion identification is proposed. Maximum Intensity Block Code (MIBC) is extracted as features. The experiments were carried out using Weizmann action dataset considering nine activities viz (walking, running, jumping, side, bend, waving one hand, waving both hands, jump-in-place-on-two-legs or pjump and skip) and the various tree based classifier like Random Forest, Naive Bayes, Random Tree, and Decision Tree (J48) utilized. In the experimental results, random forest classifier showed the best performance with an overall accuracy rate of 95.3%which outperforms other algorithms.