Oxford Spires Dataset
Yifu Tao, Miguel Ángel Muñoz-Bañón, Lintong Zhang, Jiahao Wang, Lanke Frank Tarimo Fu, Maurice Fallon
We present the Oxford Spires Dataset, captured in and around well-known landmarks in Oxford using a custom-built multi-sensor perception unit as well as a millimetre-accurate map from a terrestrial LiDAR scanner (TLS). The perception unit includes three global shutter colour cameras, an automotive 3D LiDAR scanner, and an inertial sensor — all precisely calibrated.
To download the dataset, please go to the Download section. The code is available on Github. You can find our paper from Arxiv.
Handheld Perception Unit
Our perception unit, Frontier, has three cameras, an IMU, and a LiDAR. It is shown in the figure below. The three colour fisheye cameras face forward, left, and right. The LiDAR was mounted on top of the cameras. In the table below, we show the specifications for each sensor, while in the next table, we show the ROS topics provided for each sensor.
| Sensor | Type | Rate [Hz] | Description |
|---|---|---|---|
| Hesai QT64 | LiDAR | 10 | 64 channels, 60m max. range, 104° vertical FoV |
| Alphasense Core Development Kit | IMU | 400 | Cellphone-grade, synchronised with the cameras |
| Alphasense Core Development Kit | 3 cameras | 20 | Colour fisheye, 126°×92.4° FoV, resolution 1440 × 1080, 36° overlap, synchronised with IMU |
| Topic | Rate [Hz] | Description |
|---|---|---|
/hesai/pandar |
10 | LiDAR pointclouds |
/alphasense_driver_ros/imu |
400 | IMU measurements |
/alphasense_driver_ros/cam0/color/image/compressed |
20 | Front camera images |
/alphasense_driver_ros/cam1/color/image/compressed |
20 | Left camera images |
/alphasense_driver_ros/cam2/color/image/compressed |
20 | Right camera images |
Calibration: The camera intrinsics and camera-IMU extrinsics were calibrated with Kalibr (Furgale et al., 2013). The camera-LiDAR extrinsics were calibrated with DiffCal (Fu et al., 2023). All of the intrinsic and extrinsic sensor calibration parameters are available in the dataset. LiDAR overlay in the images using this calibration is shown above.
Dataset Recording
The above-described Frontier device was mounted in a backpack (figure below). We carried this backpack through the different sites to record the sequences described in the Sites and Sequence section.
Ground Truth
For ground-truthing, we used a Leica RTC360 TLS (figure below, left). It has a maximum range of 130 m and a Field-of-View of 360° × 300°. The final 3D point accuracy is 1.9 mm at 10 m and 5.3 mm at 40 m. The point clouds are coloured using 432 mega-pixel images captured by three cameras.
We scanned the dataset’s sites from different static locations (figure below, right). From each scan, we obtained a colourised 3D point cloud.
3D reference model (TLS map): We provide the individual scans recorded as commented before (figure above, right) for each site. Moreover, we provide a merged version for each site (1cm resolution), where the scans are registered using Leica’s Cyclone REGISTER 360 Plus software. The average cloud-to-cloud error in our sites ranges from 3 to 7 mm.
Trajectory GT: The ground truth trajectory is computed by ICP registering each Frontier’s undistorted LiDAR point cloud to the TLS map described before. We do this similarly for Newer College (Ramezani et al. 2020b) and Hilti-2022 (Zhang et al. 2022). The accuracy of the ground truth trajectory is approximately 1-2 cm.
Sites and Sequence
The table below provides information about the data recorded in the different sites. We provide the dates, the number of sequences recorded and the sum of the lengths for all sequences in each site. Additionally, we provide information about if the site contains indoor parts in sequences.
| Site | Date | Sequences | Length (km) | Out-In |
|---|---|---|---|---|
| Bodleian Library | 2024-03-15 2024-05-20 2024-10-29 | 2 | 1.29 | Outdoor |
| Blenheim Palace | 2024-03-14 | 5 | 2.18 | Outdoor-indoor |
| Christ Church College | 2024-03-18 2024-03-20 | 6 | 4.12 | Outdoor-indoor |
| Keble College | 2024-03-12 | 5 | 2.87 | Outdoor-indoor |
| Radcliffe Observatory Quarter | 2024-03-13 | 2 | 0.79 | Outdoor |
| New College | 2024-07-09 | 4 | 1.66 | Outdoor-indoor |
Folder structure
In the figure below, we show the folder structure of the dataset. The data from the sensors described before is located in the raw folder. Apart from the raw, we provide processed data from VILENS-SLAM, including the undistorted point clouds, and COLMAP including the processed images. The trajectory folder contains those trajectories in TUM format. The ground truth for TLS map and trajectory is marked in red in the figure.
Code: Github
Download
Contact
We encourage you to pose any issue in Github Issues, but you can also contact us via email.