Qualitative Prediction and Recognition of ongoing Human Action Sequences Using Deep Neural Networks

Abstract

Human action detection (HAD) is an important research area in the recent decades that can be applied in many applications such as security, gaming, virtual reality interfaces, surveillance systems, etc. The authors have used object detection algorithms with deep neural network-based classifiers in this experiment. Further, the proposed research work aims to explore the solutions for object detection in HAD. The model has been provided with a set of images, wherein each image, a person will be performing an activity such as standing, sitting, bending, waving, or sleeping. The label of an image will be the activity that is being performed in those images. The model will learn this relationship, and then it can predict the label of an input that it has never seen. The model is trained and tested with a dataset consisting of random images downloaded from the internet.