Vital Sign Detection and Human Activity Recognition on 140 GHz Radar with Deep Learning Techniques

Radar-based vital sign detection and human activity recognition have gained significant attention in recent years in healthcare and security domains. The use of radar for monitoring vital signs such as heart rate, respiration rate, and activity recognition has been widely studied in the millimeter-wave frequency range, such as 24 GHz, 60 GHz, and 77 GHz. At imec, we have developed an advanced D-band (140 GHz) radar system that provides a much wider bandwidth and extremely small antenna sizes, allowing for more elements in a MIMO array and thereby achieving higher angular resolution while remaining suitable for embedding in small devices or compact spaces.

However, for vital sign detection tasks, increasing the carrier frequency poses challenges. Specifically, traditional algorithms become much more sensitive to vibrations at higher frequencies. These vibrations can degrade the performance of the system, reducing the accuracy of vital sign detection.

Moreover, the separation of human motions with breathing or heart beats, remains a significant challenge in state-of-the-art radar systems. Deep neural networks (DNNs) have gained popularity due to their ability to investigate hidden patterns in complex non-linear models. This makes them particularly suited for tackling the challenges of vital sign detection and human activity recognition.

This thesis aims to explore the use of the 140 GHz radar system to simultaneously monitor vital signs and recognize human activities using deep learning techniques. By leveraging the strengths of deep learning, we seek to address the challenges posed by the high carrier frequency, particularly its sensitivity to vibrations, thereby improving the accuracy and robustness of both vital sign detection and activity recognition.

The successful candidate should have good knowledge of signal processing, linear algebra, and machine learning. Proficiency with Matlab or python is a must. Some knowledge of radar concepts and deep neural networks is a plus.

Type:

• Master Thesis (6 months)
• Preceded by optional summer internship (3 months)

Responsible scientist(s):

• Ruoyu Feng (Ruoyu.Feng@imec.be)
• Marc Bauduin(Marc.Bauduin@imec.be)