New Trends in Machine Learning Techniques for Human Activity Recognition Using Multimodal Sensors

Abstract

The ageing of today’s society, according to demographic and epidemiological data, presents a significant increase in the elderly population. Following the experienced pandemic, research in telemedicine to improve the lives of elderly people through a comprehensive program developed by multidisciplinary teams has become a top priority. This enables the provision of remote healthcare services, facilitating access to specialists, disease monitoring, medication management, and health indicator tracking to address the medical, social, and emotional needs of elderly individuals. This study proposes a sensor-based approach to identify activity patterns without prior labels. The system architecture responsible for collecting data from the monitored user in the assisted living facility consists of a beacon, multiple anchors, and various sensors for motion, opening and closing, temperature, and humidity. The experimentation was carried out with distinct activities such as sleeping, eating, taking medication, walking, showering, and brushing teeth, inferred from the identified patterns. This approach offers an automatic and objective way to understand the routines and behaviours of older individuals, thereby improving their care and attention through personalized interventions tailored to their individual needs. Furthermore, it lays the groundwork for future research on the detection and monitoring of changes in activities over time, identifying possible signs of impairment or changes in the health of elderly people.