Human Instancing and Tracking on Sparse mmWave Radar Data

The need for non-invasive human movement monitoring in healthcare is growing, and traditional methods often involve wearable or use video monitoring. Wearable technology can cause discomfort, and video monitoring has privacy concerns. Our proposed approach uses a Millimeter-Wave (mmWave) Radar to detect and quantify human movement for health applications. One of the challenges in using mmWave radar for real-time detection is processing the range and velocity data readings. The filtered data is often sparse, which is difficult for human instancing and tracking. Our approach uses a machine learning approach with Similarity Group Proposal Network (SGPN) to detect human instancing and tracking using mmWave radar data. We show we can achieve a 94% accuracy in distinguishing between environmental noise and people, though more work is needed to track individual people.