Emergency Monitoring of OlderPeople Using Multiple Sensors with Privacy in Consideration

As the proportion of older people in the population increases, automated monitoring systems are implemented to support the caregivers and the health care systems. However, the users of these systems can feel uncomfortable as they are under constant surveillance since a monitoring system requires the collection of user data. Privacy and cybersecurity are key concerns related to the monitoring of older people. To reduce the risk of these threats, sensors apart from visual surveillance has been applied in monitoring. This project explores the use of non-invasive sensors such as a thermal camera, an Ultra-wideband radar and heart rate sensors and their efficiency in monitoring the human health state with machine learning. The sensor data is collected and preprocessed from a multimodal human activity database and used with machine learning models for performance comparison. The applied models categorise the data samples into conditions: normal, abnormal and emergency and perform prediction of heart rates based on multivariate time-series data to evaluate the capability to detect an emergency before it occurs. Separating human conditions based on limited feature information is challenging and even more if deployed in a real-time system where more variance and unexpected occurrences can happen. Machine learning is used in measuring the performance of ambient sensors. The experiments show that the performance of non-intrusive ambient sensors performs better compared to the traditional vision-based surveillance camera. The ambient sensor thermal camera contains rich information regarding health state and can provide additional security if implemented and possibly offer a reliable monitoring system while maintaining privacy