A Review of Multidata Human Activity Recognition

Human Activity Recognition (HAR) has become one of the most known and prominent research area in a variety of sectors during the last few decades. As a result, HAR can be used to keep tabs on a wide range of activities and the environment in general, all according to the user’s choices. HAR aims to categorize human actions in order to better understand how people behave. Its adaptability has recently piqued the curiosity of computer vision experts. RGB, skeleton, depth, infrared, point cloud, event stream, auditory, acceleration, radar, sensors, and Wi-Fi signals are just a few of the data techniques that can be used to describe human actions; each storing a slightly different collection of potentially relevant but separate information and each with their own set of benefits depending on the context of use. As a result, numerous previous articles attempted to investigate HAR methods through various means. The proposed method classifies the intelligence methods to HAR according to the various modalities from which input data can be collected and provide an in-depth examination of their relative merits and shortcomings. In this study, a comprehensive overview of current state of art in artificial intelligence is provided by including device modalities, data modalities, obstacles, and applications in HAR feature direction. The study provides some thoughts and proposals for future research based on the comparison of many HAR benchmark datasets.