A Spatial Millimetre-Wave Radar Point Cloud Dataset for Health Applications

This work presents a controlled millimetre-wave radar point-cloud dataset for analysing guided daily and rehabilitation-related movements in standardised indoor settings. The dataset is designed to support radar-based human motion analysis: radar point clouds serve as the primary sensing modality, while the Azure Kinect provides 3D skeleton labels for supervised learning and validation. Data were collected from 26 healthy, able-bodied adults aged 25-57 years. Participants followed reference videos to perform a set of daily and rehabilitation-related movements under two visual conditions: normal lighting and low-light settings. After quality control, the release contains approximately 3.47 minutes of retained movement data per participant per lighting condition. The original recording-level data include 265,487 radar frames and 322,833 Kinect-derived skeleton-label frames, while the processed aligned data contain 137,717 paired radar-skeleton frames. Each aligned radar frame is annotated with an action label, subject ID, and corresponding Kinect-derived 3D skeleton. By combining radar point clouds with aligned skeleton labels across controlled lighting conditions, the dataset supports tasks such as radar-based action classification, identity recognition, and keypoint detection.