Human Action Recognition for Pose-based Attention: Methods on the Framework of Image Processing and Deep Learning

Abstract

This paper presents an overview of some approaches of Human action recognition (HAR) for pose-based attention. The paper focus is on algorithms that use video processing on a given dataset. A list of the best HAR datasets is given in order to show the variety of the available videos online. Local and Global feature extraction are reviewed. Also some of the most common Deep Learning methods are studied: Recurrent Neural Network (RNN), Convolutional Neural Network (CNN) and Generative Adversarial Network (GAN). All of the methods are directed to recognise the pose and the focus of the person in a recording.