Enhancing Human Action Recognition using Machine Learning Techniques

Abstract

Human Action Recognition (HAR) represents an understanding of actions performed newlineby a human in a video. Action(s) modeling involves mapping action to a label that newlinedescribes an instance of that action. Different agents (i.e., humans) can perform newlineactions under varying speed, lighting conditions, and diverse viewpoints. Moreover, newlinefully-automated HAR systems have several challenges such as background clutter, newlineocclusion, scale, and appearance. newlineFurther, there has been increasing interest in the analysis of 3D data due to newline newlineadvancements in sensing technology. 3D data acquisition is made easier by cost- newlineeffective devices. Thereby, in a scene, information about depth can be obtained from newline newlinevarious objects and persons in the form of depth images. Moreover, 3D position of newlinebody joints is also available in the form of skeleton data. In this thesis, we address newlinethe recognition of actions from a sequence of depth maps and 3D skeleton data. newline newlineTo address issues and challenges of HAR such as background clutter and il- newlinelumination invariance, in this thesis, we have proposed different techniques. In newline newlineour proposed framework, we have used an improved learning algorithm named Self- newlineadaptive Differential Evolution-Extreme Learning Machine (SaDE-ELM) for action newline newlineclassification. In the proposed approach, we have used Depth Motion Maps-Local newline newlineBinary Pattern (DMM-LBP) for feature extraction and SaDE-ELM for action clas- newlinesification. Later, an approach is proposed to improve performance of action recog- newlinenition for depth-based input with Single Layer Feed-forward Network (SLFN) using newline newlineSelf-adaptive Differential Evolution with knowledge-based control parameter-Extreme newlineLearning Machine (SKPDE-ELM). Further, an approach is proposed for depth-based newlinedata using various feature combinations. For action classification, we have used newlineKernel-based Extreme Learning Machine (KELM) classifier. Moreover, to reduce newlinemisclassification, majority voting-based technique is applied. newline newlineRecent advancements in computer vision field using neural networks have re- newlinesulted in d