An Automated Human Action Recognition and Classification Framework Using Deep Learning

Human activity recognition has captivated the interest of researchers due to its significant applications such as smart home health care systems in which this technology can be applied to enhance the patients' rehabilitation. Different sensors can be used in a variety of ways to recognize human activity in a smartly managed environment. It is also used in pedestrian detection, robotics and human-computer interface, etc. Hence, with the development of Artificial Intelligence, researchers are keen to solve problems related to human action recognition and classification. We propose a novel method using Deep Learning (DL) algorithm for the task of human action recognition. The suggested framework is trained and evaluated on a publically available database containing recorded movements performed by both male and female participants. We tested various DL architectures and their parameters by changing epochs, learning rates, batch size, and optimizers before reaching the final architecture. The optimal architecture consists is trained on 6 epochs, a mini-batch of 128 on an adam optimizer, and a 0.001 learning rate. The system attained the highest accuracy of 98% on unseen test samples. The results prove the method's robustness and can be deployed for real-time human activity recognition and classification.