A Multi-Modal UWB-Radar-Camera Dataset for Vulnerable-Road-User Sensing at Urban Intersections

Kurzfassung

Vulnerable road users (VRUs) such as pedestrians and cyclists account for a large share of traffic fatalities, especially at urban intersections where line‑of‑sight (LoS) is often obstructed. To foster reproducible research on radio‑frequency‑based perception, we present a publicly available, fully synchronized, annotated multi‑modal dataset acquired at an intersection in the outskirts of Munich, Germany, within the VIDETEC‑2 project. The measurement campaign comprises three complementary sensing subsystems: A dense mesh of 15 ultra‑wideband (UWB) transceivers that periodically exchange sensing messages and record the full channel impulse response (CIR) for every link. Five FMCW radars mounted on two gantries, delivering range‑Doppler maps. A network of eight high‑definition (HD) cameras covering the intersection from different viewpoints, providing calibrated images, 3‑D bounding boxes, and object class labels. In addition, a GNSS‑RTK unit on a test VRU records ground‑truth position and velocity. The released package is available online and contains the data along with tools for parsing, loading, and visualizing the data. By making this richly annotated, outdoor, traffic‑realistic dataset openly accessible, we aim to accelerate research on RF-based technologies such as FMCW radar and UWB-based joint communication‑and‑sensing (JCAS) for VRU protection. The dataset can be accessed on Zenodo.