Human activity recognition in video benchmarks: A survey

Abstract

Vision-based Human activity recognition is becoming a trendy area of research due to its broad application such as security and surveillance, human–computer interactions, patients monitoring system, and robotics. For the recognition of human activity various approaches have been developed and to test the performance on these video datasets. Hence, the objective of this survey paper is to outline the different video datasets and highlights their merits and demerits under practical considerations. We have categorized these datasets into two part. The first part consists two-dimensional (2D-RGB) datasets and the second part has three-dimensional (3D-RGB) datasets. The most prominent challenges involved in these datasets are occlusions, illumination variation, view variation, annotation, and fusion of modalities. The key specification of these datasets are resolutions, frame rate, actions/actors, background, and application domain. All specifications, challenges involved, and the comparison made in tabular form. We have also presented the state-of-the-art algorithms that give the highest accuracy on these datasets.