Deep Learning Techniques for Human Abnormal Activity Recognition using Video Feature Extraction and Analysis: A Systematic Review

Abstract

In recent times, deep learning techniques has shown its ability to apply in any field including
speech recognition, image/video processing, natural language processing, and many more real-life problems
solving. On the other side Human activity recognition (HAR) has become a popular topic in research due to
its broad application for scientists and engineers. The researchers started working on the new ideas by
integrating the two emerging areas to solve HAR problems with the implementation of deep learning. The
present paper has reviewed some of the literatures related to existing HAR systems that are used to
recognize the human activities with prominence on detecting abnormal behaviors. First, a review is
conducted to understand how the deep features can be extracted and examined acquired from the video
frames. Second, a review is conducted to understand the process of analysis to understand the human
activities. Third, a review is conducted to understand how the existing HAR models or systems are
contributing to the efficient recognition of human activities in real-time scenarios. Forth, the complexities
while implementing the HAR systems are represented as challenges and mentioned finally the future scope
for analysis of input videos for abnormalities detection in human behaviour. The review of literature is to
permit the researchers to learn new techniques of deep learning while identifying the activities by enhancing
the systems performance accuracy.