Publication

Abstract : Human activity recognition (HAR) is the designated term for the automated distinguishing of physical activities conducted by individuals. The current study analyzes convolutional neural networks (CNN) in human activity recognition, particularly on datasets including images. By analyzing and correctly identifying the activities performed by individuals in image datasets, analysts and programmers can achieve a valuable understanding of human behaviors across different circumstances. HAR involves the automation of detection and categorization of activities that are performed by individuals and are included in image datasets using deep learning methodologies and advanced image processing techniques. Here, we propose a deep learning methodology that can accurately classify the given input images into various activities like sitting, standing, walking, and sleeping. To determine the model’s efficiency, we have employed a comprehensive dataset that is available in Kaggle’s repository, namely the activity dataset. This dataset offers a wide range of authentic and varied activities performed by groups of individuals across different locations. It contains an array of labeled images capturing their activities, like sitting, standing, walking, and sleeping, across diverse circumstances. This research established an unusual approach for HAR, which was empowered by CNN Drive by FastAI and offered an easy and accessible deep learning platform for users.