IPNL Intelligent Positioning & Navigation Lab Search EN 简 繁

Hong Kong UrbanNav: An open-source multisensory dataset for benchmarking urban navigation algorithms

Hsu, L. T., Huang, F., Ng, H. F., Zhang, G., Zhong, Y., Bai, X., Wen, W.

NAVIGATION: Journal of the Institute of Navigation (2023)

journal Q1 Featured page
DOI Repository Video
Overview montage of the UrbanNav dataset, including Hong Kong urban scenes, vehicle sensors, LiDAR point clouds, GNSS sky-view imagery, and skymasks.
Figure 1 from the UrbanNav paper: multisensory data and challenging Hong Kong urban scenarios for navigation benchmarking.

Summary

Open multisensory GNSS, LiDAR, camera, and IMU benchmark captured in Hong Kong urban canyons.

Videos

Figures

The dataset story starts from the physical collection platform that carries GNSS, IMU, LiDAR, and camera sensors together.

UrbanNav data collection vehicle and roof-mounted sensor kit with GNSS, IMU, LiDAR, and cameras.
Figure 2: data collection platform and integrated multisensor kit.

This turns the sensor platform into a synchronized logging pipeline for collecting aligned multisensor data.

Hardware connection flowchart for synchronized UrbanNav sensor logging and postprocessing.
Figure 3: hardware connection and synchronization flow for the data collection platform.

The installation geometry provides the calibration basis needed to use the different sensor streams together.

Sensor installation drawing with rack configuration, side view, and top view.
Figure 4: calibrated sensor installation geometry for the UrbanNav vehicle.

After the platform is defined, the paper introduces the urban scenarios that the dataset is designed to benchmark.

UrbanNav scenario trajectories and environment reconstructions for middle-class, deep, harsh, and tunnel urban cases.
Figure 5: representative trajectories and reconstructed environments for the four selected datasets.

The skymasks quantify how those scenarios obstruct satellites, connecting environment severity to GNSS difficulty.

Skymask examples for middle-class urban, deep urban, and harsh urban scenarios.
Figure 6: skymask examples used to describe satellite visibility constraints in urban canyons.

The benchmark section then runs visual-inertial and SLAM algorithms against ground truth across the selected scenarios.

Benchmark trajectories comparing ground truth with VINS-Mono, VINS-Fusion, and ORB-SLAM3.
Figure 16: benchmark trajectories comparing visual-inertial and SLAM methods against ground truth.

The final figure connects the public dataset release to practical lessons from collection, synchronization, and community reuse.

Lessons learned montage for UrbanNav, including sensor interference, start point, timing synchronization, and GitHub usage map.
Figure 18: lessons learned from collecting and releasing the UrbanNav dataset.

Key idea. A high-quality open-source multi-sensor dataset (GNSS, LiDAR, camera, IMU) captured in Hong Kong’s deep urban canyons, with accurate ground truth — purpose-built for benchmarking urban navigation algorithms.

Impact. UrbanNav has become a widely used community benchmark for urban GNSS and sensor-fusion research, lowering the barrier for groups worldwide to test in genuinely challenging conditions. Community-facing infrastructure, not just a paper.

Related project and resources

Project: UrbanNav Resource: UrbanNav Dataset