A Dataset for Victim Detection in Search-and-Rescue Operations Using Robot-Mounted UWB-Radar Sensors

Search-and-Rescue (SAR) operations rely on sensor platforms to detect victims behind walls, obstacles etc. Conventional sensors (thermal, optical etc.) face limitations in occluded environments, whereas Ultra-Wideband (UWB) radar can penetrate obstacles, offering a promising solution for victim detection. Meanwhile, robotic platforms can significantly enhance disaster response operations, especially in hazardous environments. This paper introduces a comprehensive dataset of victim measurements from a robot-mounted radar sensor. The dataset design closely emulates SAR scenarios inside buildings for victim detection and localization. To the best of our knowledge, this is the first dataset capturing robot movement during SAR operations and focusing on UWB radar sensing from multiple orientations. To demonstrate the technical validity of the proposed dataset alongside its applicability in supporting further development of Artificial Intelligence (AI) and Machine Learning (ML) models, we also present a CNN model and an XGBoost model for victim detection, achieving F1-scores of 78% and 83% respectively. Moreover, a rule-based position estimation method achieved a Mean Absolute Error (MAE) of 0.49 m, further highlighting the dataset's value.