Human action recognition approaches with video datasets—A survey

Abstract

Human Activity Recognition has recently attracted considerable attention. This has been triggered by the rapid development of advance technologies and learning methods. Human action recognition can be actively used in a number of application domains which may positively influence various aspects of the daily life. These include, (1) preventing dangerous activities and detection of crimes such as theft, murder, and property damage, and (2) predicting pedestrian activities in traffic, among others. To better serve these applications and the like, it is essential to highlight the various aspects related to the existing methods so that their actual users could realize and identify the good performing methods that work fast and are capable of recognizing the correct activities with high accuracy. The latter scope is covered in this survey which summarizes and analyzes the methods that perform learning and analysis processes on video datasets to grasp a new perspective on human action recognition. The survey also covers the major datasets commonly used in human activity recognition research. Accordingly, this survey could be recognized as a valuable source for researchers and practitioners.