IPqM-Fall: Multi-Sensor Wearable Dataset for Military Activities and Falls

OverviewThe IPqM-Fall dataset is a multi-sensor wearable Inertial Measurement Unit (IMU) dataset designed for Human Activity Recognition (HAR) in tactical, military, and survival environments. The dataset was created as part of the "Future Combatant" project by the Navy Research Institute (IPqM), with support from the Federal Center for Technological Education Celso Suckow da Fonseca (CEFET-RJ) and funding from the Funding Agency for Studies and Projects (Finep). The dataset captures the unique biomechanical signatures of subjects performing complex movements, transitions, and falls while armed (carrying a rifle) versus unarmed. Because the collection includes standard activities (e.g., walking, sitting, running, jumping), this dataset serves as a highly robust resource for both military and civilian contexts. Data was collected from 15 military personnel of varying ranks from the Navy Research Institute. To ensure high-quality and realistic tactical data, all participants had to be actively qualified for armed service with at least one year of operational experience. Detailed information—including rank, gender, age, height, and weight—was recorded for each volunteer to ensure a representative sample of the military population and to allow researchers to study biomechanical variances across different body types. Data Collection and Structure Data was collected using off-the-shelf devices equipped on each volunteer to ensure realistic form factors. Chest: LG Velvet smartphone (placed in the uniform's chest pocket). Wrists: Two Samsung Galaxy Watch 4 smartwatches (left and right wrists). The devices captured Linear Acceleration and Angular Velocity at the highest possible hardware rates (SENSOR_DELAY_FASTEST). To ensure perfect temporal synchronization across all three devices, the continuous data was resampled to a uniform 90 Hz. The raw data files are provided in .parquet format and contain the following time-series features: 3-axis Linear Acceleration (ax, ay, az) 3-axis Angular Velocity (wx, wy, wz) Signal Magnitudes (amag, wmag) The raw data represents continuous trial recordings, categorized into three operational domains: Activities of Daily Living (ADL): Continuous recordings of static postures (Standing, Sitting), locomotion (Walking, Running, Jumping), and incline navigation (Stairs, Uphill, Downhill). Military Operations (MO): Tactical sweeps, crawling, and explosive postural transitions (e.g., dropping rapidly from a run to a prone shooting position). Falls: Frontal, Backward, and Lateral falls. Armed vs. Unarmed: The majority of the ADL and Fall trials were executed in both unarmed states (hands free) and armed states (holding a rifle, resulting in altered biomechanics). The raw dataset is organized hierarchically to separate the continuous sensor signals by subject and device placement. IPqM-Fall/ └── raw/ ├── [subject_id]/ │ ├── [sensor_position]/ │ │ ├── [activity]/ │ │ │ ├── [subject_id]_[sensor_position]_[activity]_[trial].parquet │ │ │ └── ... Windowing & Labeling for Machine Learning This dataset contains unsegmented, continuous time-series signals. Raw data must be preprocessed into fixed-size windows and mapped to specific categorical labels. To ensure optimal performance, proper synchronization, and accurate alignment of impact events, it is highly recommended that you use the data-processing pipeline provided in our GitHub repository. It contains the exact methodologies, algorithms, and automated scripts required to: Segment the raw continuous files into synchronized 2-second windows (180 samples) with a 1-second stride. Apply algorithmic signal refinement (such as peak-impact isolation for falls and settling-point detection for transitions). Map the raw data to a highly structured, multi-task hierarchical labeling schema designed for Edge-AI environments. Usage and Acknowledgment This dataset was developed in association with the Brazilian Navy Research Institute (IPqM). If you utilize this raw dataset or our preprocessing pipeline in your research, please cite our accompanying paper and acknowledge its contribution to tactical biomechanics and wearable intelligence. Ethics Statement The data collection process was approved by the Ethics Committee of the Naval Hospital Marcílio Dias (Protocol CAAE 75570623.3.0000.5256) via Plataforma Brasil. Written informed consent was obtained from all military volunteers.