RadarMoCap-Posture: A Synchronized mmWave Point-Cloud and Motion-Capture Benchmark for Human Posture Recognition

RadarMoCap-Posture: A Synchronized mmWave Point-Cloud and Motion-Capture Benchmark for Human Posture Recognition RadarMoCap-Posture is a controlled testbed dataset containing ceiling-mounted 3D mmWave radar point-cloud recordings together with temporally aligned markerless motion-capture reference data for human posture and motion-state recognition. The dataset was collected in a furnished laboratory environment designed to replicate a nursing-home residential care setting. A single Texas Instruments IWR6843 mmWave radar was mounted near the centre of the ceiling, while five Captury cameras provided markerless skeletal motion-capture observations. Radar and motion-capture data were acquired concurrently, with recorded timing information provided to support frame-level temporal alignment. The dataset includes recordings from 13 adult participants (10 male and 3 female), aged 21–65 years and with heights ranging from 153–196 cm. It contains both single-person and multi-person recordings, including two- and three-person scenarios. Eleven participants have individual single-person recordings, while two additional participants (P12 and P13) are represented only in a multi-person recording. The controlled protocol covers lying, sitting, standing, and walking states. Lying, sitting, and standing include both stationary and natural-movement phases, while walking consists of free movement within the recording room. The dataset therefore provides variation in posture, movement, participant height, occupancy, and multi-person sensing conditions. Released data include radar point-cloud CSV files, radar timing/metadata files, and Captury skeletal or motion-capture export files available for the corresponding recording sessions. Original camera video is not included in the public release to protect participant privacy and in accordance with applicable data-protection and institutional requirements. The dataset is intended to support research on radar-based posture classification, motion-state recognition, temporal synchronization, radar point-cloud processing, multi-person sensing, and privacy-conscious human monitoring. The dataset is associated with the manuscript “Interpretable and Privacy-Conscious Continuous Posture Monitoring in Elderly Care Using a Single Ceiling-Mounted 3D mmWave Radar.” For complete dataset structure, acquisition details, file formats, synchronization procedures, participant information, limitations, and usage guidance, please refer to the accompanying README.