Search for a command to run...
This dataset is accompanied by the paper “"ICDUWB-26" Dataset for Infant In-Car Detection Using Ultra-Wideband Radar”, which is currently under review and will be linked after the review process is completed. The paper provides detailed information about the dataset structure, measurement scenarios, and initial benchmark results. The dataset is designed for infant detection in in-car scenarios using four monostatic ultra-wideband (UWB) radar devices. Measurements were conducted using an infant dummy capable of simulating limb, head, and respiratory movements, including breathing irregularities. Data was collected under multiple conditions: an empty car, the dummy alone, an adult alone, and both the dummy and an adult inside the vehicle. For the empty-car and dummy-only scenarios, additional measurements were performed with individuals either moving or standing next to the vehicle. Various movement patterns of all occupants were considered. Recordings were conducted using two different cars. In total, the dataset comprises 56 distinct scenarios for both cars. Measurements are performed with a target sampling frequency of 10Hz. The dataset consists of two main folders. The “recordings” folder contains one .h5 file per recording. Each file is structured into two subfolders. The “data” folder includes the radar maps (real and imaginary part) of shape NxM, with N being the number of channel impulse responses and M the number of range bins. It also includes a reference respiration signal of the adult, if present. The “labels” folder contains annotations describing the scenario, such as seating positions and movement patterns. It also includes the primary binary label, where 0 indicates that no unattended child is present in the vehicle and 1 indicates that an unattended child is present. The “datasets” folder contains two preprocessed versions of the data that were used for the benchmark experiments.