A Study on Vision-Based Human Activity Recognition Approaches

The objective of human activity recognition (HAR) is to categorize actions from subject behavior and environmental factors. Systems for automatically identifying and analyzing human activities make use of data collected from many types of sensors. Despite the fact that numerous in-depth review articles on general HAR themes have already been published, the area requires ongoing updates due to the developing technology and multidisciplinary nature. This study makes an effort to recapitulate the development of HAR from computer vision standpoint. HAR tasks are significantly associated to the majority of computer vision applications, including surveillance, security, virtual reality, and smart home. The improvements of cutting-edge activity recognition techniques are highlighted in this review, particularly for the activity representation and classification approaches. Research timeline is organized on the basis of representation techniques. We discuss a number of widely used approaches for classification and adhere on the category of discriminative, template-oriented, and generative models. This study also focuses on the major drawbacks and potential solutions.