Radar micro-Doppler for healthcare applications

The context of this research is the use of radar systems for indoor monitoring of people aimed at detecting patterns of life to infer information regarding the health status of vulnerable people such as elderly and people with cognitive or physical disabilities. Radar can provide a contactless and unobtrusive tool to perform this monitoring, with no need of using invasive cameras or wearable devices.

Micro-Doppler indicates the small modulations on the radar echoes generated by movements and rotations of various parts of a moving target, and in case of humans these are related to the small swinging movements of limbs and torso oscillations while an individual is moving. Human micro-Doppler signatures extracted from radar sensors have been investigated in recent years for a variety of applications, such as classification of different activities performed by people (walking, running, carrying objects) recognition of individuals based on their walking gait, and healthcare/assistive purposes (e.g. early detection of fall events involving elderly people, or gait characterisation in the presence of walking assistive devices). Several signal processing methods have been proposed in the literature to characterise human micro-Doppler signatures for healthcare, as well as many different features that can be extracted and processed by machine learning algorithms to perform automatic recognition and classification.

In this PhD project the candidate will perform a mix of experimental work collecting data using radar systems in realistic healthcare and indoor monitoring scenarios, as well as analysis of these data investigating different time-frequency distributions to characterise micro-Doppler signatures, different feature extraction and selection techniques, and different algorithms based on machine learning to optimise the classification performance. During this project the candidate is expected to develop the expertise to lead his/her experimental research project, to become familiar with both hardware and signal processing aspects related to radar & sensing, and to interact effectively with colleagues from different backgrounds and disciplines.

There is significant scope for a detailed investigation of many aspects and approaches to improve the overall system performance, in particular the identification and selection of the most suitable features and the fusion of information from multiple radar sensors or from different types of sensors, which are expected to be more and more present in indoor environments with the development of Internet of Things technologies. There is interest in characterising the normal walking gait prior to a fall event, especially when assistive walking devices such as canes/walkers are used, in reducing false alarms related to actions such as picking up objects from the floor, and in testing algorithms in realistic experimental indoor scenarios with multiple sensors. There is interest in applying novel deep learning techniques for improved classification of indoor movements and patterns of life.

These reliable monitoring systems can be beneficial not only for fall detection, but also to evaluate more generically the pattern of life of an individual, for instance how active the person is, how often he/she moves in different part of the house and what activities are performed, in particular fundamental activities such as food intake. Irregularities with respect to the normal pattern of life of a person can be used for early detection of deteriorating health conditions (for instance initial symptoms of dementia), providing the opportunity for timely and more effective treatment.

This project is believed to fit with the EPSRC 'Healthy nation' objective, in particular the ambitions 'transform community health & care' and 'improve prevention & public health'.