Title: FRED: A Multi-Modal Autonomous Driving Dataset for Flooded Road Environments

Abstract

To the best of our knowledge, the Flooded Road Environments Dataset (FRED) stands as the inaugural multi-modal autonomous driving dataset designed specifically for scenarios involving water-related hazards. The collection comprises imagery captured by a 2.3 MP FLIR Blackfly USB3 camera, 64-beam 360-degree point clouds generated by an Ouster OS1-64 LiDAR, and inertial measurement unit (IMU) data from an iXblue ATLANS-C, which was refined using Geoflex RTK GNSS corrections. These diverse data streams were gathered from five distinct sites, recorded during and subsequent to flooding occurrences.

To facilitate usability, the dataset is available in two formats: KITTI-style, ensuring seamless compatibility with standard data processing tools, and RTMaps, which allows for the direct playback of the vehicle’s data acquisition. Semantic annotations are included to support the training and assessment of both single-sensor and sensor-fusion approaches for identifying water hazards. Furthermore, the inclusion of positional and velocity data, along with recordings taken under dry weather conditions, aids in the development of map-integrated, location-based detection algorithms and enables the evaluation of additional tasks such as localization and SLAM.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC