Human activity recognition and behavioural prediction: a comprehensive systematic review

Recent developments in computer vision, deep learning, machine learning, and other related fields have increased the demand for Human Activity Recognition (HAR) studies. This paper offers an organized analysis of the current techniques, emphasizing machine learning and deep learning methods applied to human activity recognition and the prediction of behaviours. The goal of this comprehensive review is to analyse the development and future prospects of 3D human pose estimation, with implications for a broad range of applications. In this study we collected scientific papers from IEEE Xplore, ACM Digital Library, Springer Link, and ScienceDirect. Papers including machine learning and deep learning approaches were included in the review, and the study concentrated more on deep learning techniques for HAR. By linking action attributes like talking, eating, walking, running, sitting, and sleeping, we can construct a model for human behaviour. However, anticipating the next actions poses a challenge. A prediction method uses numerous approaches that are described later in this paper to identify the probable future behaviour or state by exploiting the characteristics of human behaviour. These methodologies fall under the domain of Human Activity Prediction (HAP), frequently producing a value derived from a probabilistic change in state from past observations of the individual’s behaviours and the interconnectedness of different states. Human Activity Recognition (HAR) identifies and classifies physical activities performed by individuals using sensor data or video analysis techniques. Human Action Prediction (HAP) is the process of forecasting future human actions based on observed behavioural patterns and contextual information. Their interdependence is significant as precise recognition improves prediction models, and strong predictive capabilities can, in turn, enhance recognition accuracy, particularly in complex or unpredictable environments. In addition, we discuss the challenges in HAR and behavioural prediction and highlights emerging technologies for prospective development. Our findings emphasize the need to combine the domains of HAR and HAP to improve the accuracy of outcomes that are futuristic.