Download Report IEEE DataPort : Edge-Efficient Two-Stream Multimodal Architecture for Non-Intrusive Bathroom Fall Detection - 2026
AVI, CSV, TXT by anonymity anonymity
Information
Format: AVI, CSV, TXT Publisher: IEEE DataPort Publication Date of the Electronic Edition: 01/15/2026
?
ISBN: 10.21227/xx7n-v439
$15
$3Discount Coupon
Delivery time: Maximum 24 hours
Description
This dataset contains synchronized millimeter-wave (mmWave) radar and floor vibration recordings collected in a full-scale bathroom mock-up for research on non-intrusive fall detection. The aim is to support the design and evaluation of privacy-preserving monitoring systems that operate in wet bathroom environments where cameras and wearable devices are difficult to use.The recording environment is a tiled bathroom with a raised shower platform, standard fixtures and running water. Two sensing nodes are installed. The first node is a compact mmWave radar mounted on the upper wall at 1.87 m height, providing line-of-sight coverage of the shower area and the surrounding floor. The second node is a triaxial accelerometer rigidly attached to the shower platform to capture floor vibration at impact. Both nodes are driven by microcontroller-class hardware with synchronized timestamps and transmit buffered windows to an edge gateway.Eight background scenarios are included: empty bathroom, light object drop, heavy object drop, upright walking, flexed-torso walking, wall-supported walking, quiet standing and squatting. Within these scenarios, intentional falls are interleaved so that every background condition contains both fall and non-fall examples. All human activities are performed with the shower running so that the acoustic and mechanical conditions reflect realistic bathroom use.The radar node produces sparse three-dimensional point clouds at 12.5 Hz. Raw radar frames are stored as per-frame PLY files and as per-frame kinematic descriptors in comma-separated value files. The vibration node samples three-axis acceleration at 200 Hz and the dataset provides time-stamped triaxial acceleration series resampled to 100 Hz. In total the release contains more than 3 hours of recordings, approximately 1.1×10⁵ radar frames and 3.1×10⁶ vibration samples. Frame-level labels indicate fall or non-fall, scenario type, water state and fall direction. Subject-independent splits with 60 percent of subjects for training, 20 percent for validation and 20 percent for testing are included so that users can perform reproducible benchmarking of vibration-only, radar-only and multimodal radar–vibration fall detection methods under edge-computing constraints.
$15
$3Discount Coupon
Delivery time: Maximum 24 hours