underwater-datasets

Multi Camera Underwater Visual Inertial Dataset

Data moved to Hugging Face (October 2026). NTNU Box has been discontinued, so all download links below now point to huggingface.co/datasets/ntnu-arl/underwater-datasets. Each trajectory is now a single merged .bag, verified message-by-message against the original split recordings. The dataset page on Hugging Face has the full documentation with measured durations and sizes.

About

This is an underwater dataset consisting of multiple onboard sensors, mainly, 5 Camera and IMU ( Alphasense Core Research Developent Kit),a Pressure Sensor, and actuator commands from an ROV piloted in an indoor facility (MC-lab). A total of 24 trajectories are collected with following characteristics:

• 4 Motion patterns with length 100-120m
• External illumination ranging from 100 to 700 lux
• Onboard illumination at 3 gradual levels

Additionally, we provide trajectories with varying media (Air, Water).

Updates / News

• Oct 17 2026: NTNU Box links replaced by Hugging Face links. All data, calibration and MoCap ground truth is now hosted at huggingface.co/datasets/ntnu-arl/underwater-datasets.
• May 27 2024: Added Mocap files for subset-3 in (.tum) format with artifacts removed.
• Apr 7 2024: Added 6 trajectrories (2 lighing conditions x 3 motion patterns) with peicewise-continuous ground-truth from motion capture. Link: Subset 3
• Nov 29 2023: Uploaded undertwater dataset with multi-camera imu system for underwater VIO and camera models. Link: Subset 1 and Subset 2

Varying illumination at 6 levels

5 Camera and IMU

Cam0	Cam1	Cam2	Cam3	Cam4

Ariel: Underwater robot with multi-camera imu and compute suite

Robot in water	System Description

Motion Patterns

Shape	Distance Travelled (m)	Mean Linear Velocity (m/s)	Mean Angular Velocity (rad/s)
Rectangular	120	0.25	0.12
Infinity	108	0.45	0.16
Dumb-bell	110	0.45	0.20
Dumb-bell w/ Transverse Motion	110	0.25	0.11

Trajectories with Visual Inetrial Data

Subset 1: Motion Patterns

The first subset contains 24 trajectories with 4 motion patterns and varying illumination.

Each of the trajectory contains following data in ROS bag format:

1. Camera | fps: 20Hz | resolution: 540p*720p | format: Mono8 | Topics:

/alphasense_driver_ros/cam0  # Front Stereo Pair: Right
/alphasense_driver_ros/cam1  # Front Stereo Pair: Left
/alphasense_driver_ros/cam2  # Top Camera
/alphasense_driver_ros/cam3  # Right Side Camera
/alphasense_driver_ros/cam4  # Left Side Camera

2. IMU (synced with cameras) | fps: 200Hz | Topics:

/alphasense_driver_ros/imu

3. IMU on FCU (Pixhawk) | fps: 50Hz | Topics:

/mavros/imu/data

4. Pressure Sensor | fps: 50Hz | Topics:

/mavros/imu/static_pressure

5. Actuator output PWM | fps: 50Hz | Topics:

/mavros/rc/out # ch 1-8 thrusters, ch 9-10 for Lumen LED

The collective data can be found (1-18) and (19-24), whereas the following table contains link to each trajectory:

Motion Pattern 1 (Rectangular)

Shape	Distance Travelled (m)	Mean Linear Velocity (m/s)	Mean Angular Velocity (rad/s)
Rectangular	120	0.25	0.12

---

No.	Illumination	Duration (s)	Size (Gb)	Link
1	E0+O1	460	18.1	link
2	E0+O2	456	17.8	link
3	E0+O3	392	15.3	link
4	E1+O3	386	15.1	link
5	E1+O2	443	17.3	link
6	E1+O1	395	15.4	link

Motion Pattern 2 (Infinity)

Shape	Distance Travelled (m)	Mean Linear Velocity (m/s)	Mean Angular Velocity (rad/s)
Infinity	108	0.45	0.16

---

No.	Illumination	Duration (s)	Size (Gb)	Link
7	E1+O1	251	9.8	link
8	E1+O2	252	9.8	link
9	E1+O3	262	10.2	link
10	E0+O3	281	11.0	link
11	E0+O2	279	10.9	link
12	E0+O1	263	10.2	link

Motion Pattern 2 (Dumbbell)

Shape	Distance Travelled (m)	Mean Linear Velocity (m/s)	Mean Angular Velocity (rad/s)
Dumb-bell	110	0.45	0.20

---

No.	Illumination	Duration (s)	Size (Gb)	Link
13	E0+O1	242	9.8	link
14	E0+O2	271	9.8	link
15	E0+O3	138	10.2	link
16	E1+O3	318	11.0	link
17	E1+O2	294	10.9	link
18	E1+O1	290	10.2	link

Motion Pattern 2 (Dumbbell w/ Transverse Motion)

Shape	Distance Travelled (m)	Mean Linear Velocity (m/s)	Mean Angular Velocity (rad/s)
Dumbbell w/ Transverse Motion	110	0.25	0.11

---

No.	Illumination	Duration (s)	Size (Gb)	Link
19	E1+O1	362	9.8	link
20	E1+O2	316	9.8	link
21	E1+O3	365	10.2	link
22	E0+O3	409	11.0	link
23	E0+O2	415	10.9	link
24	E0+O1	485	10.2	link

Illumination Details: External light has two levels E0 corresponding to 100 - 300 lux and E1 corresponding to 400-700 lux measured outside and above water surface. Onboard light varies between three incremental levels, O1, O2, and O3, corresponinding to 20%, 40%, 60% pwm level of the Lumen Subsea LED (Max: 1500 lumens) respectively.

Subset 2: Trajectories with Varying Media (Air, Water)

The second subset contains 6 trajectories with varying media (Air, Water).

Trajectory	Duration	Size	Link
1	173	2.6	link
2	126	1.9	link

Camera IMU Calibration

The cameras were calibrated using the Kalibr toolbox. The calibration data is provided in the following table with the links for each camera intrinsics and extrinsics.

Camera	Intrinsics (Air)	Intrinsics (Water)	Extrinsics
cam0, cam1(Stereo Pair)	link	link	link
cam3	link	link	link
cam4	link	link	link

Subset 3: Trajectories with Ground-truth from Motion Capture

No.	Illumination	Duration (s)	Size (Gb)	Motion Pattern	Link	MoCap Ground-Truth (.tum)
1	E1+O1	370	14.5	Rectangular	link	link
2	E1+O1	374	14.7	Infinity/Figure “8”	link	link
3	E1+O1	318	12.5	Dumbbell	link	link
4	E0+O1	405	15.9	Rectangular	link	link
5	E0+O1	401	15.7	Infinity/Figure “8”	link	link
6	E0+O1	302	11.6	Dumbbell	link	link

Reference

If you use this work in your research, please cite the publications as follows:

An Online Self-calibrating Refractive Camera Model with Application to Underwater Odometry

@misc{singh2023online,
      title={An Online Self-calibrating Refractive Camera Model with Application to Underwater Odometry}, 
      author={Mohit Singh and Mihir Dharmadhikari and Kostas Alexis},
      year={2023},
      eprint={2310.16658},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}

Contact

You can contact us for any question:

• Mohit Singh
• Mihir Dharmadhikari
• Kostas Alexis