UrbanNav Dataset

The UrbanNav project provides several datasets collected in various Hong Kong urban environments. These datasets include ROSBAG files, GNSS RINEX files, IMU data, and Ground Truth.

The Hong Kong datasets were collected using a Honda Fit platform equipped with a specific sensor suite. This information is useful for understanding the data modalities and hardware constraints when developing localization algorithms.

Sensor Suite:

• 3D LiDAR: HDL 32E Velodyne (360° HFOV, +10°~-30° VFOV, 80m range, 10Hz)
• Slant LiDARs: VLP16 Velodyne or Lslidar C16
• IMU: Xsens Mti 10 (400 Hz, AHRS)
• GNSS Receivers: u-blox ZED-F9P (x2), EVK-M8T (x1) at 1 Hz
• NovAtel Flexpak6: 1 Hz
• Camera: ZED2 Stereo (15 Hz)
• SPAN-CPT: RTK GNSS/INS (RMSE: 5cm, 1Hz)

The dataset uses GPS time (weeks and seconds since Jan 6, 1980) and ROS time (Unix epoch). To synchronize them, you must account for the offset between the GPS epoch and the Unix epoch, as well as leap seconds.

For data collected in May 2021, the GPS-UTC offset is 18 leap seconds.

Conversion Logic:

1. Calculate total GPS seconds: (weeks * 604800) + seconds.
2. Add the epoch offset: 315964800 seconds.
3. Subtract leap seconds: 18.

Example calculation for week 2158, second 95593: (2158 * 604800 + 95593) + 315964800 - 18 = 1621218775 (UTC).

function utctime = gps2utc(gps_week, gps_seconds)
    SECONDS_IN_GPS_WEEK = 604800.0; 
    utctime = (gps_week * SECONDS_IN_GPS_WEEK + gps_seconds-18) + 315964800  ; % 18 leap seconds
end

If the projection is inaccurate due to rotation errors, you can adjust the R matrix in the camera_to_imu extrinsics:

1. Convert the existing R matrix to Euler angles using an external converter (e.g., Rotation Converter).
2. Modify the Euler angles to correct the orientation.
3. Convert the new Euler angles back into an R matrix to update your configuration.

To perform Visual Inertial Odometry using the VINS-Fusion framework on UrbanNav data, follow these steps:

1. Download the dataset .bag file.
2. Clone the VINS-Fusion repository.
3. Update the camera intrinsics and extrinsics in your specific .yaml configuration file.
4. Launch the ROS nodes and play the bag file using the following commands:

roslaunch vins vins_rviz.launch
rosrun vins vins_node ~/catkin_ws/src/VINS-Fusion/config/vi_car/vi_car.yaml
rosbag play DATASET.bag

The Tokyo dataset (UrbanNav-TK-20181219) contains approximately 4.14 GB of data covering over 10 km of path length.

Note on Calibration: LiDAR calibration files and extrinsic parameters between sensors are not available for this dataset. For GNSS/LiDAR/IMU integration studies, it is recommended to use the Hong Kong dataset instead, though the Tokyo GNSS data is specifically useful for studying challenging urban canyon environments.

Download Links:

• Dropbox Link
• Baidu Clouds Links (Password: 7xpo) (Recommended for users in mainland China)

The lidar2image tool is used to project LiDAR points into the image plane. This is useful for estimating or verifying camera_to_imu extrinsics by converting points through the following transformations: LiDAR frame $\rightarrow$ Body frame $\rightarrow$ Camera frame $\rightarrow$ Image plane.

To build the tool, use the standard CMake workflow:

mkdir build && cd build
cmake ..
make
./lidar2image

If you are using raw GroundTruth files (e.g., UrbanNav_TST_GT_raw.txt), you must convert the coordinates to the ENU (East-North-Up) frame.

Conversion tools are located in the repository at: UrbanNavDataset/tools/gt_vis/.

The UrbanNav dataset provides GNSS measurements in RINEX format. To simplify data access and processing, the authors provide an open-source package called GraphGNSSLib.

Capabilities of GraphGNSSLib:

• Data Access: Provides easy access to GNSS RINEX files.
• ROS Integration: Publishes GNSS data as customized ROS messages.
• Positioning Algorithms: Supports GNSS positioning and Real-Time Kinematic (RTK) using Factor Graph Optimization (FGO).

https://github.com/weisongwen/GraphGNSSLib