Intelligent Human Action Recognition: A Framework of Optimal Features Selection based on Euclidean Distance and Strong Correlation

Abstract

Extracting salient and most prominent features from a given video sequence is one of the most critical steps in human action recognition.  In this article, proposed a novel method for human action recognition, which efficiently addresses the problem of selection of most prominent and robust features in the feature selection step. Three types of features are fused based on their highest values and later most optimal features are selected with an implementation of a novel Euclidean distance (ED) and strong correlation (SC) method. In the final phase, selected features are classified using multi-class support vector machine (M-SVM). Four publically available datasets are utilized including Weizmann, KTH, UCF YouTube, and HMDB51 with an improved classification accuracy of average more than 94%. Experimental results authenticate our claim that the proposed method outperforms compared to several existing methods.