AN ADVANCED REVIEW ON HUMAN ACTIVITY RECOGNITION USING ARTIFICIAL INTELLIGENCE

Understanding human interactions and relationships requires human activity recognition (HAR).Extraction is difficult since it reveals a person's identity, personality, and mental state.Human activity movement forecasting is HAR.HAR is used in many applications since cellphones and video cameras can collect human activity data.Digital devices, apps, and AI advances have made deep learning data extraction for reliable detection and interpretation possible.This integration has strengthened our understanding of HAR's three pillars-acquisition devices, artificial intelligence, and their applications-which are expanding rapidly.While various review papers have covered the basics of HAR, few have compared all HAR devices and even fewer have examined the impact different AI designs.This assessment covers 2006-2021 and analyses HAR's three pillars.The study also suggests ways to improve HAR design for reliability and stability.Five main conclusions: (1) HAR is built on devices, AI, and apps; (2) HAR dominates the healthcare business; (3) hybrid AI models are still developing and require significant improvement to provide stable and reliable designs.Additionally, these models must have reliable predictions, high precision, effective generalisation, and the ability to achieve application objectives without bias; there has been little research on anomaly detection in human activities; and movement prediction has advanced little.The three core components of the HAR industry-electronic devices, apps, and artificial intelligence-will continue to evolve, and AI will shape the future of the sector.This paper summarises current human activity classification research.We classify human activity recognition methods and evaluate their pros and cons.Two main types of human activity classification techniques use one or more modalities.Subcategories within each category show how they represent human activities and which ones they highlight.We also exhaustively evaluate publically accessible human activity recognition datasets and ideal HAR dataset criteria.This paper highlights sensor-based and video-based human activity detection development, emphasizing on core technologies, identification systems, and applications from low-level to highlevel representations.