Deep-learning-based human activity recognition for Alzheimer's patients' daily life activities assistance

Alzheimer's disease is considered as one of the most well-known illnesses in the elderly. It is a neurodegenerative and irreversible brain disorder that slowly destroys memory, thinking ability, and ultimately the ability to perform even basic daily tasks. In fact, people suffering from this disorder have difficulty remembering events, recognizing objects and faces, remembering the meaning of words, and developing judgment. As a result, their cognitive abilities are impaired and they are unable to perform activities of daily living independently. Therefore, patients need constant support to carry out their daily activities. In this study, we propose a new support system to support patients with Alzheimer's disease to carry out their daily tasks independently. The proposed assistance systems are composed of two parts. The first is a human activity recognition (HAR) module to monitor the patient behaviour. Here, we proposed two HAR systems. The first is based on 2D skeleton data and convolution neural network, and the second is based on 3D skeleton and transformers. The second part of the assistance systems consists of a support module that recognizes the patient's behavioural abnormalities and issues appropriate warnings. Here, we also proposed two methods. The first is based on a simple conditional structure, and the second is based on a reinforcement learning technique. As a result, we obtain four different assistance systems for Alzheimer's patients. Finally, a comparative study between the four systems was carried out in terms of performance and time complexity using the DemCare dataset.