Artificial Intelligence Enabled Methods for Human Action Recognition using Surveillance Videos

Abstract

Computer vision applications have been attracting researchers and academia. It is more so with cloud computing resources enabling such applications. Analysing video surveillance applications became an important research area due to its widespread applications. For instance, CCTV camera are used in public places in order to monitor situations, identify any theft or crime instances. In presence of thousands of such surveillance videos streaming simultaneously, manual analysis is very tedious and time consuming task. There is need for automated approach for analysis and giving notifications or findings to officers concerned. It is very useful to police and investigation agencies to ascertain facts, recover evidences and even exploit digital forensics. In this context, this paper throws light on different methods of human action recognition (HAR) using machine learning (ML) and deep learning (DL) that come under Artificial Intelligence (AI). It also reviews methods on privacy preserving action recognition and Generative Adversarial Networks (GANs). This paper also provides different datasets being used for human action recognition research besides giving an account of research gaps that help in pursuing further research in the area of human action recognition.