Development of Deep Neural Framework for Human Activity Recognition

In the domain of deep learning and computer vision, Human Activity Recognition (HAR) holds paramount importance in understanding human movements through sensor data. This research presents a robust HAR system leveraging the VGG-19 deep learning architecture. Acknowledging challenges like limited context understanding and dataset biases, the research emphasizes the ongoing need for mitigation through continued research. Despite these hurdles, the VGG-19 based system showcased promising applications in healthcare and smart environments, enriching human-machine interactions. This research work conducted a comparative analysis between VGG-19 and ResNet-50 using standardized images of 15 human activities, highlighting VGG-19’s superiority with an impressive 98.44% accuracy on the test dataset. The research underscores the potential of deep learning, particularly VGG-19, in accurately identifying intricate human activities from images. However, challenges such as contextual understanding and biases persist, prompting the research to emphasize the ongoing efforts to enhance the reliability and ethical soundness of HAR systems. These advancements are crucial for diverse sectors like healthcare and surveillance, ensuring the effective and responsible deployment of HAR technology.