A hybrid elephant herding optimization and support vector machines for human behavior identification

Human behavior identification has a great importance in daily life. Security, gaming, and medical diagnosis are vital applications in that field. This paper introduces a hybrid classification approach for human behavior identification employing support vector machines (SVMs) classifier hybrid with Elephant Herding Optimization algorithm (EHO). The Elephant Herding Optimization algorithm used to fine-tune SVM parameters and to select most discriminant features. Validation of the proposed approach will be accomplished using a computer vision-based data set named Vicon. It was acquired from multiple human action detection experiments. Results show superiority for the proposed approach over other techniques on the same data set regarding classification accuracy. EHO-SVM hybrid algorithm reaches 91.21 % and 90.62 % accuracies for two test cases with different action class selections.

I. Introduction

This paper introduces a rich topic where machine learning and optimization techniques are devoted for the purpose of human behavior and action identification. Classification of human actions has been an active area of research as it involves numerous applications in our daily life. Human activity and behavior analysis can be accomplished with the aid of behavioral information sources such as images, video feeds, and sensors [1]–[3]. Actions identification has been applied for numerous important applications such as medical diagnosis [4], security, and entertainment. Most surveillance systems depend on vision-based techniques, and they appear everywhere now in the streets, stores, and organizations [5], [6]. The detection of abnormal behaviors is the main focus of surveillance systems whether computerized or using the aid of human operators to take proper real-time response [7].