2026-10-14T10:28:30Z https://www.scidb.cn/oai

10.57760/sciencedb.0130w 2026-10-08T11:09:27Z

2026-10-08 Huangpu River Water–Air Integrated Multimodal Perception Dataset doi:10.57760/sciencedb.0130w en This dataset was developed by the Wireless AI research team at Shanghai Jiao Tong University, focusing on the integrated research of intelligent perception, wireless propagation modeling, and communication perception in the Huangpu River urban waterfront environment. The dataset is based on Unreal Engine 5 to construct a high fidelity 3D simulation scene of the Huangpu River in Shanghai, and combined with ASVSim to achieve motion, hydrodynamics, and multi-sensor simulation of fixed shore base stations and ships; The water air collaborative scenario framework supports further integration with unmanned aerial vehicle platforms. When generating data, first configure the base station, initial pose of the ship, camera, LiDAR, and Radar parameters through settings. json, and select encounter or chase motion scenes, as well as environmental conditions such as sunny days, rainfall, clouds, fog, snow, and different time periods. After the automatic acquisition program starts the simulator, the scene state is recorded with a unified logical frame number: RGB, Depth, GPS, pose and other visual/state modes are exported according to the main frame, and LiDAR and Radar export point clouds according to their sampling frequency. Subsequently, based on the pose of the base station and the moving target in the same frame, Sionna RT is used to perform electromagnetic ray tracking using the XML scene of the Shanghai Jiang'an Corridor calibrated with coordinates, generating multi band complex CSI and its propagation path parameters. CSI preserves the original complex channel coefficients without normalization, facilitating subsequent channel modeling, path analysis, and comparison across different frequency bands. The data is generated by a Windows GPU workstation, with main tools including Unreal Engine 5, ASVSim, Python, Open3D/NumPy, and NVIDIA Sionna RT. The scene geometry and materials are converted into XML format that Sionna can read; During the collection phase, sensor observations, target status, and frame level metadata are synchronously saved to ensure that different modalities can be read based on ship identification and the same frame number.  Huangpu River; wireless sensing; multimodal fusion; Synthetic Data Zhong Haofeng Mo Jianhua PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank