Classifying Physical Actions of Human Models using Multi-objective Clustering based on Elephant Herding Algorithm

Abstract

In artificial intelligence robot design understanding the behavior of humans play a significant role. The human behavior can be aggressive or normal which is characterized by its arm, leg and body moments. The problem to identify the human behavior is dealt here as a multi-objective clustering problem where the segregation is carried out using two objective functions: intra cluster distance and inter cluster distance. A new multi-objective clustering algorithm ‘MOEHA’ is proposed based on recently reported Elephant Herding Algorithm (EHA). The MOEHA performance on clustering is demonstrated on three synthetic and six real world datasets. Comparative results demonstrated superior performance over benchmark algorithms NSGA-II and MOPSO. Five case studies on the classification of human physical actions demonstrated better results over the comparative algorithms.