Human Activity Recognition using GRU Algorithm

Abstract

The recognition of human activity is a challenging task nowadays, where surveillance cameras are employed for security purposes and monitoring. It is highly appreciated to utilize the best deep learning techniques to achieve higher accuracy than existing methods. In this research, the Gated Recurrent Unit (GRU) technique is employed to identify activities performed by humans. The Human Activity Recognition Trondheim (HARTH) dataset is utilized, which consists of data collected from individuals. The dataset comprises six different classes of daily activities performed by humans: climbing downstairs, upstairs, walking, sitting, running, and standing. The algorithm is implemented against the HARTH dataset to achieve higher accuracy using TensorFlow and the Python framework, and accuracy is calculated. A confusion matrix is also obtained from the conducted research. This research concludes that the GRU algorithm yields a higher accuracy of around 95% in identifying human activities compared to the machine learning algorithms implemented earlier.